fix: Handle DST transitions in group_by_dynamic (#25410)

[wtn]

Fixes #25410.

group_by_dynamic panicked when window boundaries fell on non-existent datetimes during DST transitions, even when the actual data didn't contain those times.

Reproduction

import polars as pl
import datetime as dt

df = pl.DataFrame({
    "timestamp": pl.datetime_range(
        dt.datetime(2024, 2, 7, 9, 31),
        dt.datetime(2024, 2, 24, 16, 0),
        time_zone="America/New_York",
        eager=True,
        interval="1m"
    )
}).join(pl.DataFrame({"id": [1, 2, 3]}), how="cross")

df.group_by_dynamic(
    "timestamp",
    every="1m",
    period=dt.timedelta(days=17),
    closed="both",
    label="left",
    group_by="id"
).agg(ts_last=pl.col("timestamp").last())
# Panics: datetime '2024-03-10 02:00:00' is non-existent in time zone 'America/New_York'

Cause

The BoundsIter in polars-time calls .unwrap() on duration addition when computing window boundaries. With a 17-day period starting from Feb 7, the window boundaries include March 10, 2024 2:00 AM—a non-existent time in America/New_York when DST springs forward to 3:00 AM. This triggers try_localize_datetime to raise an error with NonExistent::Raise.

Fix

Added fallback helper functions that catch non-existent datetime errors during window boundary calculations. When a boundary falls on a DST gap, the code skips forward by 1 hour and retries. This maintains performance in the common case (just a Result check) while gracefully handling DST transitions.

[codecov]

Codecov Report

❌ Patch coverage is 91.25964% with 34 lines in your changes missing coverage. Please review.
✅ Project coverage is 79.70%. Comparing base (8ab1657) to head (a3832da).

Files with missing lines	Patch %	Lines
crates/polars-time/src/windows/window.rs	81.50%	27 Missing ⚠️
crates/polars-time/src/windows/test.rs	98.22%	4 Missing ⚠️
crates/polars-time/src/windows/group_by.rs	83.33%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #25459      +/-   ##
==========================================
+ Coverage   79.58%   79.70%   +0.11%     
==========================================
  Files        1743     1743              
  Lines      240439   240779     +340     
  Branches     3038     3038              
==========================================
+ Hits       191347   191903     +556     
+ Misses      48310    48094     -216     
  Partials      782      782              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[orlp]

I feel this is a bit dangerous. We don't check if the error is indeed a DST transition, and even if true I'm not sure if skipping a full hour is always the correct behavior.

[wtn]

@orlp Applied your feedback. The fix now uses a fallback helper that catches non-existent datetime errors during window boundary calculations and skips forward past the DST gap when a boundary falls on it.