Evaluate FFmpeg large-file handling for the provider router

[WalksWithASwagger]

Context

Legacy whisperforge contains useful large-file strategy notes around FFmpeg validation, ffprobe inspection, chunking, parallel transcription, transcript assembly, and fallback behavior. Evaluate those ideas against the current provider matrix and canonical run model.

Audit: docs/AUDIO-REPO-CONSOLIDATION-AUDIT-2026-05-18.md

Acceptance Criteria

• Document whether FFmpeg chunking should become part of the canonical transcription router.
• Identify which backends need large-file support, timestamps, diarization, streaming, and privacy fixtures.
• Define file-size thresholds, chunking strategy, fallback behavior, and temp-file cleanup expectations.
• Add or specify fixture coverage before changing runtime defaults.
• Do not copy the old whisperforge monolith or Supabase assumptions.

Verification

• python3 -m json.tool ops/roadmap/features.json
• git diff --check
• Any prototype must also run make test.

[WalksWithASwagger]

Shipped in 2a7e658. Added docs/LARGE-FILE-ROUTER-EVALUATION-2026-05-19.md and linked it from the provider matrix/consolidation audit. Runtime defaults are unchanged; the next step is fixture-backed router capability metadata before FFmpeg behavior changes. Verification passed: python3 -m json.tool ops/roadmap/features.json; git diff --check; make test; make eval-fixture; venv/bin/python tests/ui_smoke.py; make smoke.