BUG — GLOWS: instrument team exclusion-handling

[leowerneck]

Change Summary

Closes #3474.

Overview

GLOWS L1b exclusions defined by the instrument team are keyed by a unique block identifier: the UTC start time of an L1b block. The instrument team and the SDC sometimes use different sclk kernel versions when converting this time, which can shift the identifier by a couple of seconds (up to ~1 s observed so far). The current exact string comparison in flag_from_mask_dataset() misses those shifted entries and silently drops the corresponding exclusions.

This PR replaces the exact match with a timestamp comparison using a 5-second tolerance, picking the closest entry when more than one falls within the window. A 5-second tolerance is safe because the shortest L1b block lasts ~15 s. The change matches the suggested implementation by @mstrumik in #3474.

File changes

• imap_processing/glows/l1b/glows_l1b_data.py: In HistogramL1B.flag_from_mask_dataset(), match block identifiers as datetime64[s] within a ±5 s tolerance instead of by exact string equality, selecting the closest entry on multiple matches.
• imap_processing/tests/glows/test_glows_l1b_data.py: Add test_flag_from_mask_dataset_tolerance and test_flag_from_mask_dataset_picks_closest_within_tolerance covering exact match, within-tolerance, boundary, beyond-tolerance, and closest-of-several cases.

Testing

Running the GLOWS test suite:

uv run pytest imap_processing/tests/glows

results in 96 out of 96 tests passing.

[lacoak21]

Nice ! Looks good.