rtcp: bound per-SSRC MOS history without losing lifetime aggregates (#2174)

[darwvin-dev]

Fork-local draft for upstream #2174.

Changes

• Cap per-SSRC RTCP history at 512 samples.
• Preserve lifetime MOS average/count and independent extrema snapshots when the history is trimmed.
• Extract thread-safe ssrc_stats_record() so the same production path is unit-testable.
• Add a deterministic 700-report test to t/test-stats.c that verifies queue size, lifetime MOS/jitter totals, discarded duplicate timestamp handling, and extrema timestamps from entries already evicted.
• Keep NG query/delete stats and teardown logs on lifetime counts.
• Clarify that detailed MOS progression only describes recent samples.

Validation / upstream readiness

• Review upstream implementation and commit changes on a separate fork branch.
• Add deterministic >512-sample unit coverage (t/test-stats.c).
• Add RTCP output-shape assertions (t/auto-daemon-tests-rtcp.pl).
• Run make asan-check; GitHub Actions are currently not running on this fork.
• Run full daemon RTCP suite and benchmark RAM on realistic long calls.
• Check any consumer requiring full-lifetime MOS progression.

No test execution claimed. Keep as Draft until build/ASAN passes.

---

Upstream submission notes

This is a draft, carried from contributor fork PR darwvin-dev/rtpengine#2.
Unfinished validation: Please refer to the outstanding build, runtime, regression and/or integration tests above. No claim is made that these checks passed. Review for compatibility and completeness before marking ready.
The source branch is behind current upstream master and may require reconciliation before merging.

[rfuchs]

FTR this is already pending upstream: 6bd002e

[darwvin-dev]

Thanks for pointing this out! I'll close my draft to avoid duplication Happy to contribute regression tests if useful.

[darwvin-dev]

Closing in favor of upstream 6bd002e.