Fix exiftool_xml GPS epoch times for per-sample timestamps

[ptpt]

Summary

Fixes 4 failing integration tests where the exiftool_xml / exiftool_runtime geotag source produced GPS tracks that diverged from the native source.

Cameras like the GoPro MAX store a single GPSDateTime per sample, shared across all (~17) GPS coordinates in that sample. The exiftool XML parser discarded these timestamps (set epoch_time to None) because the epoch-time count (1) didn't match the coordinate count (~17), while the native parser broadcasts the single sample timestamp to every coordinate.

This caused MAPGPSTrack to contain None epoch times under exiftool_xml but real timestamps under native, failing the exact-equality assertion in assert_descs_exact_equal.

Fix

In _aggregate_epoch_times, when there is exactly one GPS epoch time for multiple coordinates, broadcast it to every coordinate (matching the native parser). Genuine length mismatches (e.g. 3 vs 2) still fall back to None with a warning.

Tests

• All 52 integration tests pass (previously 4 failing):
  • test_process_geotag_with_exiftool_xml
  • test_process_geotag_with_exiftool_xml_pattern
  • test_process_geotag_everything_with_exiftool_runtime
  • test_process_video_geotag_source_with_exiftool_runtime
• Updated the unit test that locked in the old discard-on-mismatch behavior to assert the new broadcast behavior, and added a separate case for genuine (3-vs-2) mismatches that still fall back to None.
• mypy and ruff clean.