fix(timeline): normalize naive datetimes from DB to offset-aware

[rogerdigital]

Summary

• get_latest_end() and _row_to_block() could return naive datetimes when stored ISO strings lacked a TZ offset, causing TypeError on comparison with aware datetimes from _stem_to_dt() and _now() — this silently breaks the entire timeline pipeline after the first block is written
• Add defensive astimezone() normalization in all DB read paths and in _capture_stem_in_window() as a belt-and-suspenders guard

Tests

• test_get_latest_end_naive_string_returns_aware — naive ISO string → aware datetime
• test_get_latest_end_aware_string_stays_aware — aware stays aware
• test_get_latest_end_no_rows — empty table → None
• test_row_to_block_naive_datetime_becomes_aware — query_recent returns aware blocks
• test_capture_stem_in_window_naive_start_end — naive start/end don't crash, correct window membership

All 19 tests pass, ruff clean.

Closes #24

[gemini-code-assist]

Code Review

This pull request ensures that datetimes processed in the timeline aggregator and store are timezone-aware by normalizing naive datetimes using the .astimezone() method. These changes are applied to _capture_stem_in_window, get_latest_end, and _row_to_block, and are accompanied by new unit tests to verify correct handling of naive versus aware datetimes. Feedback indicates that the timestamp derived from the filename stem in the aggregator should also be normalized to prevent potential TypeError exceptions during comparison if it remains naive while the window bounds are aware.

[gemini-code-assist]

The comparison start <= ts < end on line 51 will still raise a TypeError if ts is naive and start/end are aware (or vice versa). While _stem_to_dt usually returns aware datetimes, it can return a naive one if the filename stem is exactly 20 characters and lacks a valid 'p' or 'm' offset prefix (e.g., 2026-04-21T17-07-32_). To be truly 'belt-and-suspenders' as intended by this PR, ts should also be normalized to an offset-aware datetime before the comparison.

Suggested change

	if start.tzinfo is None:
	start = start.astimezone()
	if end.tzinfo is None:
	end = end.astimezone()
	if ts.tzinfo is None:
	ts = ts.astimezone()
	if start.tzinfo is None:
	start = start.astimezone()
	if end.tzinfo is None:
	end = end.astimezone()

[rogerdigital]

Fixed in 4c74c74 — added ts normalization before comparison.

[rogerdigital]

@Xiao-ao-jiang-hu The ts normalization is already included in commit 67d212b (line 47-48 in aggregator.py). Gemini's review was on an earlier version — all three variables (ts, start, end) are normalized before comparison now. Ready for re-review.