query_bucket clips events 1 ms past the end of the query period

[ErikBjare]

Found by a Python/Rust query parity suite (ActivityWatch/activitywatch, PR to follow). aw-core 33c8528.

Bucket.get() (aw_datastore/datastore.py) rounds the end of the requested period up to the next millisecond:

milliseconds = 1 + int(endtime.microsecond / 1000)

This adds 1 ms even when endtime is already a whole millisecond, so an event that runs past the period end is clipped 1 ms after it.

Minimal repro

Insert {"timestamp": "2026-01-01T10:00:00Z", "duration": 60, "data": {}} into bucket a and query with timeperiod 2026-01-01T09:00:00Z/2026-01-01T10:00:30Z:

RETURN = query_bucket("a");

• aw-server (Python): duration: 30.001
• aw-server-rust: duration: 30.0

Any view whose period ends while an event is still running (for example "today" with a current window event) reports 1 ms too much and extends past the period. Rounding up only when there are sub-millisecond digits (math.ceil(endtime.microsecond / 1000)) would keep the "don't miss events" intent without the offset.

[ErikBjare]

Parity suite: ActivityWatch/activitywatch#1467. The affected cases are strict xfails there, so they XPASS once the fix reaches the pinned submodules.