fix(api): correct DST spring-forward handling in parse_time_range

[lntutor]

Closes #39230

Summary

libs/datetime_utils.py::parse_time_range resolved a non-existent (DST spring-forward) wall-clock time by adding a hardcoded one hour, assuming every gap is exactly one hour. Real tzdata has other gap sizes, producing two bugs:

• Gaps > 1 h (e.g. Antarctica/Troll +2 h, Pacific/Apia 24 h): +1h lands inside the gap, so the retry tz.localize(..., is_dst=None) raises NonExistentTimeError again — uncaught, leaking a raw pytz exception out of a function documented to raise only ValueError.
• Gaps < 1 h (e.g. Australia/Lord_Howe +30 min): the result is shifted by an hour instead of the actual gap, so the returned UTC instant is wrong (16:15 UTC instead of 15:45 UTC for 02:15), silently shifting analytics time-range bounds.

Change

 except pytz.NonExistentTimeError:
-    dt += datetime.timedelta(hours=1)
-    return tz.localize(dt, is_dst=None).astimezone(utc)
+    return tz.normalize(tz.localize(dt, is_dst=False)).astimezone(utc)

tz.normalize(tz.localize(dt, is_dst=False)) shifts a non-existent time forward by exactly the gap size for any offset change (30 min, 1 h, 2 h, ...), matching zoneinfo fold=0. It never re-raises. Standard one-hour gaps are unchanged (America/New_York 02:30 → 07:30 UTC). The AmbiguousTimeError (fall-back) branch is untouched.

Verified the fix matches stdlib zoneinfo fold=0 for Antarctica/Troll, Australia/Lord_Howe, Pacific/Apia, and America/New_York.

Tests

Replaces the previous mock-based test_parse_time_range_dst_nonexistent_time (which patched pytz.timezone and asserted the old localize call sequence rather than a real result — it used Jan 1, not an actual DST transition) with real-timezone tests:

• standard 1 h gap — America/New_York 02:30 → 07:30 UTC (unchanged behavior),
• gap > 1 h — Antarctica/Troll 01:30 must not raise and resolves to 01:30 UTC,
• gap < 1 h — Australia/Lord_Howe 02:15 → 2024-10-05 15:45 UTC.

All 25 test_datetime_utils tests pass; the full tests/unit_tests/libs suite stays green; ruff check and ruff format --check clean.

[github-actions]

Pyrefly Type Coverage

Metric	Base	PR	Delta
Type coverage	54.01%	54.01%	-0.00%
Strict coverage	53.50%	53.50%	-0.00%
Typed symbols	34,407	34,407	0
Untyped symbols	29,577	29,579	+2
Modules	3029	3029	0