perf(bench): complete the release fanout comparison

.github/relay-cpu-bisect.md

@@ -111,7 +111,7 @@ Against `main` at [619df6e1b3](https://github.com/moq-dev/moq/commit/619df6e1b3)
 
 `Cursor::probe` is the relay's check for groups a reader has missed. Each time a subscriber asks for its next group, the probe looks at what arrived on the track since its last look and decides whether any group the reader still wants has left the cache without the live feed delivering it (evicted, skipped, or from before the feed started), and asks upstream to fill those. A live reader that takes each group as it arrives never has such a gap, yet it paid for the whole check on every group: locking the subscription and the track's shared state, cloning the track's fills, walking the arrival ledger, checking each new group for holes, and running the ask-and-answer loop. Each time the cache aged out an old group, the eviction also landed with every reader, all of which had already taken it, and forced each one to re-run the full check. The fast path judges a group the arrival walk serves without the hole, backfill, and ledger passes, skips the subscription, fills, and answers when nothing is asked or owed, and starts each reader's eviction inbox above what it handed out. It recovers most of [moq-dev/moq#6171](https://github.com/moq-dev/moq/pull/6171)'s and [moq-dev/moq#7339](https://github.com/moq-dev/moq/pull/7339)'s cost.
 
-kio's `WaiterList::register` was the session path's second-largest self cost: 7.2k per group. A waiter recorded 8 lists, and a relay session's control and serving tasks parked on more, so every list past the eighth took a duplicate entry on every poll. Remembering 16 cut the session by 4,424 (-3.9%) and was initially reverted after two moq-net integration tests failed. [moq-dev/moq#7789](https://github.com/moq-dev/moq/pull/7789) fixes the front's metadata/abort ordering and the quiet-catalog test's detached handshake, then raises the record limit to 16; its paired session measurement is recorded in that PR. Forcing every idle gate to walk still reproduced both failures, so the suspected idle-gate wake dependency did not explain them. The per-group costs this report attributes and nothing here cut are accepted in the [release-perf epic](../quest/m1/release-perf/README.md).
+kio's `WaiterList::register` was the session path's second-largest self cost: 7.2k per group. A waiter recorded 8 lists, and a relay session's control and serving tasks parked on more, so every list past the eighth took a duplicate entry on every poll. Remembering 16 cut the session by 4,424 (-3.9%) and was initially reverted after two moq-net integration tests failed. [moq-dev/moq#7789](https://github.com/moq-dev/moq/pull/7789) fixes the front's metadata/abort ordering and the quiet-catalog test's detached handshake, then raises the record limit to 16; its paired session measurement is recorded in that PR. Forcing every idle gate to walk still reproduced both failures, so the suspected idle-gate wake dependency did not explain them. The per-group costs this report attributes and nothing here cut are accepted in the [release performance report](release-perf.md).
 
 On the quiet test host (run.we4umM, one-minute load at most 4.7), chat over io\_uring went from 154.5k \[154.1-155.0] instructions per frame on `main` to 148.9k \[148.3-150.8] with the probe and window changes (-3.6%), four rounds each, every cell at 20,000 frames/s with no loss. The relay's QUIC, crypto, and syscalls are most of its cost, so the model's -23% shows as a few percent there.
 
@@ -127,7 +127,7 @@ On the quiet test host (run.we4umM, one-minute load at most 4.7), chat over io\_
 Neither cost is in the probe's judgement, so the fast path left both in place:
 
 - **[moq-dev/moq#6995](https://github.com/moq-dev/moq/pull/6995), the live index's walk.** It indexed the live feed's groups apart from fetched backfill, so every search for live content (`TrackState::live`) walked a chain of two flattened options where it had walked the cache's range. The walk's callers are the drift edge each read measures (`poll_drift` → `drift_edge` → `live_edge`) and the arrival walk, on the reader's cursor and its forward-drop cursor alike; the judgement never calls it. It still cost 2,840 inclusive per group on `main`, as at [moq-dev/moq#7542](https://github.com/moq-dev/moq/pull/7542)'s merge (2,817). The probe's own share of the PR (+1,115 self) did fall with the fast path: the probe's self cost went from 4,343 at the merge to 1,842 on `main`.
-- **[moq-dev/moq#7542](https://github.com/moq-dev/moq/pull/7542), layered window queries.** A window answers each query from its shared fixed runs and a layer of moving ones (`runs::Layered::first` and `last`): +3,810 inclusive at the merge, less about 800 in the range-map queries they replaced. [moq-dev/moq#7605](https://github.com/moq-dev/moq/pull/7605)'s single-set fast path brought them to 2,117 on `main`, about 40 queries per delivery at about 50 instructions each, from the forward-drop walk, peeks, admission, and the probe. That is the number of window queries a per-reader window answers, not a walk, so it is accepted in the [release-perf epic](../quest/m1/release-perf/README.md).
+- **[moq-dev/moq#7542](https://github.com/moq-dev/moq/pull/7542), layered window queries.** A window answers each query from its shared fixed runs and a layer of moving ones (`runs::Layered::first` and `last`): +3,810 inclusive at the merge, less about 800 in the range-map queries they replaced. [moq-dev/moq#7605](https://github.com/moq-dev/moq/pull/7605)'s single-set fast path brought them to 2,117 on `main`, about 40 queries per delivery at about 50 instructions each, from the forward-drop walk, peeks, admission, and the probe. That is the number of window queries a per-reader window answers, not a walk, so it is accepted in the [release performance report](release-perf.md).
 
 `TrackState::live` now walks either the cache or the live index through one branch per call, and the drift edge's single-run search scans it statically rather than through a `dyn` iterator (run.m71Z8E, against `main` at [fa9c1bf2a7](https://github.com/moq-dev/moq/commit/fa9c1bf2a7)):
 
@@ -178,7 +178,7 @@ Each change is counted against the row above in its own run (N037x9, vS3CAR, DV1
 - **The budget.** A read that blocks before its group's first frame judges the group's expiry. `Consumer::poll_expiry` asked the logical reader's budget (`Recover::poll_budget`: the reader's lock and a sync of its mirrored preferences) before `is_expired`, which answers false without the budget until a group lands above. The sync is redundant there: a preference update mirrors into every copy at once (`Mirror::sync`), a route change syncs every reader (`serve_from`), and the landing that makes the budget matter wakes the read through `watch_landing`, which then asks the budget. `is_expired` now takes the `Recover` and asks only past that check.
 - **The forward drop.** The relay's track side polls the reader for runs to report (`TrackRun::poll_forward_drop`) on each track wake, and again after a receive that parked, to collect a stale skip the receive's probe queued without a wake. Each poll judged every copy (`Reader::let_go`) with nothing to report, and the second probed the live edge again (`Cursor::poll_drift` and `probe`) right after the receive had, with the same waiter. The reader now returns once nothing is queued, and the second poll only collects. Drop probes fall from 5 to 3 per delivered group.
 
-What remains of each viewer's live read is accepted in the [release-perf epic](../quest/m1/release-perf/README.md): about 3.5k per delivered group in the drop poll's probe and 3.5k in the receive's (three each, one wake of each kind), and 0.9k in the expiry checks of a read that blocks before its first frame and of a group waiting for stream credit, which park on the next landing.
+What remains of each viewer's live read is accepted in the [release performance report](release-perf.md): about 3.5k per delivered group in the drop poll's probe and 3.5k in the receive's (three each, one wake of each kind), and 0.9k in the expiry checks of a read that blocks before its first frame and of a group waiting for stream credit, which park on the next landing.
 
 ## `group_relay_frames` since `e7a5c10`