Fix fast-path executed price to match the recorded bid and quote

[AryanGodara]

Description

Fast-path orders settled on-chain at a buy amount that matched didn't match the of chain signed price. PR attempting to fix the issue.

Changes

• shared::fee::adjust_bid_for_gas_costs nets the quote's gas fee out of (sell, buy) the same way the user-facing quote does (sell orders scale buy down; buy orders raise sell), with unit tests.
• stage the quote's gas gas_cost_in_sell_token on StagedSolution (#[serde(default)] so already-staged rows keep prior behaviour).
• autopilot fast path executes finalize_bid (account for gas cost and then apply volume fees), in both the pre-flight limit check and the recorded-bid computation.
• driver finalize_fast_path_solution rebuilds the winning fulfillment with haircut_fee = 0, so it settles the autopilot's bid exactly.

How to test

• added e2e test and unit test
• updated existing e2e tests to use strict equality on executed values instead of approximate equality

[claude]

Claude finished @AryanGodara's task in 6m 34s —— View job

---

PR Review: Fix fast-path executed price

I reviewed the fast-path fee-netting change against the quoting/validation flow. Overall the core logic is sound: apply_quote_fee correctly mirrors the user-facing quote (sell orders keep the full sell and scale the buy by (sell-fee)/sell; buy orders raise the sell by fee), and applying it before the volume fee makes the recorded bid and the pre-flight limit check match what the driver settles. Nice unit tests, and the mainnet-repro test is a good touch.

I found one substantive gap, posted inline:

• crates/shared/src/quote_storage.rs:141-146 — the staged fee is FeeParameters::fee() (bare gas), but the user-facing quote and the orderbook fast-path validation both net out fee_with_additional_cost(additional_cost) (ERC-1271 sig-verification gas, plus hook gas for unverified quotes). So for non-EOA / hooked orders the recorded bid still over-shoots the quoted buy — a smaller version of the mismatch this PR fixes. EOA (EIP-712/EthSign) orders have additional_cost == 0, which is why the tests (all EOA) don't surface it. [details inline]

Non-blocking notes:

• The autopilot pre-flight now uses adjusted_sell = quoted_sell (full) while the orderbook check uses quote.sell_amount (= quoted_sell - fee). The autopilot value matches on-chain settlement, and the difference only makes the autopilot stricter (safe: order falls through to a regular auction), so no action needed — flagging for awareness since the PR's theme is orderbook/autopilot parity.
• The PR description mentions a driver finalize_fast_path_solution / haircut_fee = 0 change; that already landed on main in Replace haircut logic with existing volume fee logic #4900 and isn't part of this diff — description is fine, just noting.
  · branch aryan/fix-fast-path-executed-price

[claude]

The staged fee uses FeeParameters::fee() (bare gas_amount * gas_price / sell_token_price), but the fee the user-facing quote and the orderbook fast-path validation net out is fee_with_additional_cost(additional_cost) — i.e. it also includes ERC-1271 signature-verification gas and, for unverified quotes, hook gas.

• User-facing quote: to_final_quote → with_additional_cost(additional_cost) before with_scaled_sell_amount, so quote.buy_amount is scaled by the full fee.
• Orderbook: OrderQuoter::find_quote applies with_additional_cost(...) (order_quoting.rs:738), and check_fast_path_limit_price_fits validates against that quote.buy_amount.

Because this only nets out the bare gas fee, apply_quote_fee in the autopilot subtracts less than the real fee, so the recorded bid ends up higher than the quoted buy for any order with non-zero additional_cost (ERC-1271 / Safe orders, or unverified quotes with hooks — see the comment at order_validation.rs:1056 "Quoted gas does not include gas for ERC1271 signatures"). That partially reintroduces the exact circuit-breaker mismatch this PR is fixing, just scaled down to the additional-cost portion.

For pure-EOA (EIP-712 / EthSign) quotes additional_cost is 0 so this path is unaffected — hence the unit/e2e tests (all EOA) don't catch it.

Consider persisting the final quote.fee_amount (with additional cost) instead of recomputing .fee() here, or threading the signing scheme / additional cost into the staged competition. Note the additional cost technically depends on the order's eventual signing scheme, so exact parity may require capturing it at placement — worth confirming the intended behaviour.