fix(weather): metric visibility of 8000 m or more is VFR

[roznet]

For roznet/flyfun-weather#757 (consumer PR: roznet/flyfun-weather, branch claude/pensive-dijkstra-uqrgni).

What

A European report of 8000 m converted to 4.97 SM and landed in the FAA MVFR band (3 to 5 SM). flyfun-weather raised false "TAF at ETA: VFR → MVFR" alerts from TEMPO … 8000 SHRA. 5 SM is 8047 m: only the unit conversion made 8000 m marginal.

• WeatherReport.visibility_unit ("M" / "SM", None when unknown) is new. The parser sets it, and TAF group application, _worse_of and to_dict/from_dict carry it through.
• WeatherAnalyzer.flight_category: a metric report with visibility ≥ METRIC_VFR_VISIBILITY_M (8000, exported) is VFR. SM reports keep the FAA edge (5SM MVFR, 6SM VFR). A report with no unit falls back to the FAA rule, as before.
• This departs from the issue's suggested test. The issue said to treat "has visibility_meters" as metric. That doesn't work because the parser fills metres for SM reports too (5SM → 8046 m), so 5SM would have read VFR. The unit field is the fix for that.

Additive, no breaking API change → minor bump to 0.22.0 via /release. flyfun-weather's PR pins euro-aip>=0.22.0.

No Swift counterpart: RZFlight has no flight-category code (checked).

Verification

• pytest tests/briefing/test_weather: 337 passed. 13 are new: METAR 8000/9999/7000/5000/4000, 5SM/6SM, no-unit fallback, dict round trip, the EGJB TAF with TEMPO 8000 → VFR, TEMPO 7000 → MVFR.
• Full euro_aip suite: 1122 passed, 1 failed (test_aip_parser_parse_airports). It fails the same way on the branch without this change.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MDyEuQSS1UNjQU4AViBbxf

---

Generated by Claude Code

[claude]

Code Review

Reviewed the metric-visibility VFR fix: new visibility_unit field, parser/TAF overlay/_worse_of propagation, flight_category change, design doc and tests. The change is additive. The unit-based approach is correct, since visibility_meters is filled for SM reports too. The unit propagates through every construction path I found, and there is no Swift counterpart to update. The design doc matches the code, and the version bump is correctly left to /release. No issues found.