chore: add multicodec HEVC+AVC repro fixture and characterization test (closes #63)

[birme]

Summary

Adds a reproducing fixture and characterization test for the multicodec (AVC + HEVC) master-manifest collision described in #63.

• New fixture src/plugins/__tests__/fixtures/multicodec/ with two variants advertised at the SAME BANDWIDTH=2000000 — one CODECS="avc1.4d401f" and one CODECS="hvc1.1.6.L93.90" — plus their media playlists.
• New src/plugins/__tests__/multicodec_master.test.ts characterization test asserting the CURRENT (buggy) behavior so CI stays green: both codecs are advertised at bw 2000000, getBandwidths() collapses colliding variants to a single ['2000000'], and served segments are HEVC while the avc1 profile has no matching served segments.

Root cause seam (for #64)

The engine builds the client master from HLSVod.getUsageProfiles(); @eyevinn/hls-vodtolive keys this.segments[bw] by bandwidth only, so two variants colliding on bandwidth share one segment set and one master<bw>.m3u8. A fix must key by (bandwidth, codec) or dedupe colliding bandwidths.

PROOF

• PROOF: npm run lint → clean
• PROOF: npm run pretty → All matched files use Prettier code style!
• PROOF: npm test → Test Suites: 2 passed, 2 total / Tests: 7 passed, 7 total
• PROOF: npm run build → tsc EXIT=0

Closes #63

[birme]

pr-reviewer verdict: APPROVE

Reviewed as a distinct (non-authoring) agent. This is a deliberate REPRODUCTION/characterization test for the multicodec (AVC+HEVC) bandwidth-collision bug (#63), asserting current buggy behavior so it PASSES and keeps CI green. That is acceptable for a repro ticket; the fix is scoped to #64/#65.

Deterministic proof

• PROOF: npm run lint -> exit 0 (eslint, no errors)
• PROOF: npm run pretty -> "All matched files use Prettier code style!" (exit 0)
• PROOF: npm test -> "Test Suites: 2 passed, 2 total / Tests: 7 passed, 7 total" (exit 0); multicodec suite: 3/3 passed
• PROOF: npm run build -> tsc --project ./ exit 0

Repro is genuine (independently verified against @eyevinn/hls-vodtolive)

Ran HLSVod.load() directly against the fixture:

• getBandwidths() -> ["2000000"] (two colliding variants collapse to one key)
• getUsageProfiles() -> advertises BOTH avc1.4d401f and hvc1.1.6.L93.90 at bw 2000000
• getMediaSegments()['2000000'] -> serves hevc_720p_*.ts only

Confirms the advertised-CODECS vs served-segment mismatch is real, not a rigged assertion. The test faithfully documents the engine's actual behavior.

Checklist

1. Deterministic proof: PASS (all four commands green)
2. No commercial product names/trademarks: PASS (only standard codec/ISO-BMFF terms avc1/hvc1/HEVC/AVC; placeholder hosts mock.example, vod.dummy)
3. Matches docker-fast conventions: PASS (jest, tests layout, TS, path.join fixtures; eslint-disable is justified for the untyped JS require)
4. Security: PASS (no credentials; all file paths built from static fixture names via path.join(__dirname), no external input reaching a path/URL)
5. Commit message: PASS (chore: add multicodec HEVC+AVC repro fixture and characterization test (closes #63), conventional style)

Findings

• File: src/plugins/tests/multicodec_master.test.ts:1-2,113,138 - Severity: suggestion - Finding: Comments reference the fix as "#65" (top-of-file) while the ticket list has both #64 (trace) and #65 (honor codec preference). Minor cross-ref imprecision, not blocking. - Recommendation: When #65 lands, flip the assertions (toBeGreaterThan(0) -> toBe(0)) as the comments already instruct.
• File: src/plugins/tests/multicodec_master.test.ts:68-75 - Severity: suggestion - Finding: The mediaManifest fetcher always returns hevc_720p.m3u8 for the colliding bw, which mirrors the engine's bandwidth-only keying and is the point of the repro. Clear as written. - Recommendation: none; keep as-is.

No critical/major/minor issues. APPROVE (posted as comment since the automation account cannot self-approve).