Add fixture-backed media normalization for the transcription router

[WalksWithASwagger]

Context

The provider matrix is now documented, but large audio/video intake still needs fixture-backed media inspection and normalization before defaults change. This issue turns the matrix into a tested router/media-normalization slice while keeping current runtime defaults stable.

Acceptance Criteria

• Large audio/video inputs can be inspected and normalized through an FFmpeg-backed plan before transcription.
• Router fixtures cover local/private, cloud, chunked, timestamped, and diarization-capable paths.
• Runtime defaults remain unchanged until the router has explicit privacy and cost receipts.
• Docs connect the provider matrix to the implemented router behavior.

Tests/Evals

• Add or update focused tests in tests/test_audio.py for media inspection, normalization planning, and router capability selection.
• Use fixtures/mocks rather than live provider calls.

Verification

• make test
• venv/bin/python -m pytest tests/test_audio.py -q
• git diff --check

Agent Instructions

Roadmap ID: wf-router-media-normalization.

Work primarily in whisperforge_core/audio.py, existing audio tests, and docs/TRANSCRIPTION-PROVIDER-MATRIX-2026-05-18.md. Keep defaults unchanged. Do not require FFmpeg to be present for ordinary unit tests; mock subprocess/probe boundaries or use tiny deterministic fixtures. Make privacy/cost implications visible as data in the plan rather than prose buried in docs.

Out of Scope

• No provider default switch.
• No live transcription provider calls in tests.
• No full parallel chunk execution rewrite unless required by the tested planning layer.
• No UI redesign beyond whatever small display hook is needed for the plan.

Linear

Not linked yet. Source roadmap entries: ops/roadmap/features.json and docs/NEXT-ROUND-PLAN-2026-05-19.md.

[github-actions]

Agentic dev loop started on branch codex/PLAN-2026-add-fixture-backed-media-normalization-for-the-t with provider noop.

[github-actions]

Agentic issue quality gate failed: missing sections: Context, Tests/Evals, Agent Instructions, Out of Scope; acceptance criteria must include markdown checkboxes

[github-actions]

Agentic dev loop failed: No ready label found

[github-actions]

Agentic dev loop started on branch codex/MATRIX-2026-add-fixture-backed-media-normalization-for-the-t with provider noop.