feat(checks): add required topic inventory

[k4its1t]

Summary

• Add an evidence-only required_topics built-in check that records each requested topic's presence and message count plus missing_topic_count.
• Read the authoritative channel inventory and aggregate counts across duplicate channels, so legal multi-channel MCAP topics do not raise.
• Document the sixteenth built-in check and its configured registration pattern.

Why

HFlow could measure timing, motion, camera integrity, and content, but had no built-in answer to whether a recording contained every topic its rig was expected to produce. This adds the missing inventory evidence while leaving pass/fail policy in curation SQL.

Closes #6

Validation

• RED: uv run pytest tests/test_checks.py -q failed during collection because required_topics did not exist.
• GREEN: uv run pytest tests/test_checks.py -q — 32 passed.
• uv sync --locked --all-extras — passed.
• uv run ruff check --fix — passed; final uv run ruff check passed after rebasing onto current main.
• uv run ruff format — passed; final uv run ruff format --check reports 148 files formatted.
• uv run ty check — passed.
• uv run pytest -q — 756 passed, 6 skipped on current main.
• lychee --no-progress --include-fragments --exclude '^https://github\.com/Hebbian-Robotics/hflow/(issues|security/advisories/new)$' --exclude-path references/mcap-spec.md --exclude-path references/foxglove-CompressedVideo.proto . — 286 links checked, 0 errors.

Checklist

• I added or updated outcome-focused tests for changed business logic.
• I updated documentation for changed behavior, flags, formats, or requirements.
• I ran uv run ruff check --fix, uv run ruff format, and uv run ty check.
• I ran the relevant pytest suite.
• I did not add recordings, generated media, credentials, private URLs, or runtime artifacts.
• I preserved stored-data compatibility; built-in check changes use their own content-hash versions and do not require a transform behavior bump.

AI assistance

OpenAI Codex assisted with implementation and PR preparation. I reviewed the final diff and ran every validation command listed above locally.

[kstonekuan]

LGTM. Validated locally on current main: ruff, format, ty clean, 759 passed, 3 skipped.

Two details I checked because they are the ones that bite a new built-in: test_no_two_builtin_checks_claim_the_same_measurement_key now includes required_topics, so missing_topic_count is proven unique, and the docs table lists all sixteen rows. Aggregating over Episode.channels instead of Episode.channel(topic) is the right call, and the multi-channel test earns it.

I pushed one fixup to your branch (bb20ad8): PORTING.md:291 still said "the fifteen packaged checks".

One thing, and it is an invitation rather than a rule: the good first issue pool is best left for newcomers, and you are past that now. Higher-leverage work if you want it is running HFlow against a real corpus (Egocentric-10K or Egocentric-100K on Hugging Face are the ones we care about) and reporting what breaks, what is slow, or what is awkward. That kind of report is worth more to us than anything on the starter list.