[ramondalmau]

Summary

Fixes units audit item D2. neats_json_to_flights dropped the UTC offset of each
trajectory ts without applying it (12:02:00+02:00 became 12:02 UTC instead of 10:02),
and schema-valid RFC 3339 forms (fractional seconds, lowercase z) raised ValueError,
aborting the whole batch.

New helper _ts_to_utc_string in neats_io.py:

• applies the UTC offset;
• accepts lowercase t/z and a space separator;
• truncates fractional seconds (maintainer decision);
• reads a ts without offset as UTC (maintainer decision);
• raises ValueError naming the flight and trajectory point for invalid values.

Output contract unchanged: naive UTC string "%Y-%m-%d %H:%M:%S".
The audit report gets a follow-up status table and a status note under D2.

Not in this PR: an invalid ts still stops the whole batch (separate PR).

Evidence that nothing breaks

• Identical output on all existing data: 148 NEATS JSON files on the NEATS server
  (85,128 flights, 10,995,586 points; every ts is +0000, Z or +00:00) compared
  exactly before/after (values, dtypes, column order, attrs): 0 differences.
• Full suite with BADA and weather data: 267 passed, 4 skipped, 23 xfailed
  (baseline 231/4/23 + 36 new tests).
• Adversarial review (3 independent reviewers): differential fuzzing of 309,011
  timestamp forms against the old code and two reference parsers (0 disagreements with
  the references; no UTC input changes); all callers checked incl. Dataiku recipes;
  new tests fail on the old code (19/23). Edge cases they raised are fixed.
• ruff, ruff format, project mypy clean.

Note on the pre-commit mypy hook

The mirrors-mypy hook fails for every commit, independently of this change: its
isolated env installs numpy 2.5.4, whose stubs use Python 3.12 syntax while mypy targets
3.11. This commit was made with SKIP=mypy; project mypy (uv run mypy src/pyneats/)
passes. A separate PR will fix the hook.