fix(transform): handle zero-duration base and sub events in merge_subwatcher_fields

[fanxing11]

Found two related bugs in merge_subwatcher_fields while poking at the new transform. Both stem from how the boundary/intersection logic deals with zero-length Timeslots.

Bug 1 — zero-duration base event is silently dropped

base = [Event(timestamp=t0, duration=timedelta(0), data={'app':'x'})]
sub  = [Event(timestamp=t0-timedelta(seconds=5), duration=timedelta(seconds=10), data={'project':'P'})]
merge_subwatcher_fields(base, sub, ['project'])
# -> []   (event vanishes)

When the base is instantaneous and does overlap a sub, boundary_points collapses to a single point, so zip(boundary_points, boundary_points[1:]) is empty and nothing is emitted. The no-overlap fast path further up returns the base unchanged, so the symptom only shows up when there's a sub.

Fix: dedicated branch for instant bases that picks the active sub via the same "latest sub wins" rule and returns a single enriched zero-duration event.

Bug 2 — instantaneous sub event "colors" the whole base

base = [Event(timestamp=t0, duration=timedelta(seconds=10), data={'app':'x'})]
sub  = [Event(timestamp=t0+timedelta(seconds=3), duration=timedelta(0), data={'project':'P'})]
merge_subwatcher_fields(base, sub, ['project'])
# -> [Event(t=0, dur=10s, data={'app':'x', 'project':'P'})]

Expected: the sub is active for zero time → no slice should be enriched.

Root cause: Timeslot([0,3]).intersection(Timeslot([3,3])) returns a zero-length but truthy Timeslot. The segment loop only checked if not segment_period.intersection(sub_period): continue, so every segment adjacent to the instant matched the sub. Then the dedup pass merged the (now identical) segments back into one fully-colored event.

Fix: ignore zero-length intersections both when collecting overlapping subs for a non-instant base and inside the per-segment matcher. ip.duration > 0 is the right rule — a sub that was active for zero time can't enrich any slice. (Instant base is still handled correctly, via the dedicated branch above.)

Tests

Added three:

• test_merge_subwatcher_fields_zero_duration_base_event_preserved — covers bug 1
• test_merge_subwatcher_fields_zero_duration_base_event_no_overlap_preserved — locks in the existing no-overlap fast path so future refactors don't regress it
• test_merge_subwatcher_fields_zero_duration_sub_does_not_color_base — covers bug 2

All 30 test_transforms tests pass; ruff clean.

Real-world relevance

Most watchers heartbeat with a nonzero pulsetime, so canonical queries probably don't hit either case today. But the function is exposed in query2 and via the Python API, and instantaneous events are a perfectly valid shape (manual API posts, edge transformations, future watchers reporting markers). Wanted to lock down the contract before this transform shows up in more pipelines.

[greptile-apps]

Greptile Summary

This PR fixes zero-duration interval handling in merge_subwatcher_fields. The main changes are:

• Adds a dedicated path for zero-duration base events.
• Ignores zero-duration sub intersections for durationful base slices.
• Adds tests for instant bases, no-overlap instant bases, and instant sub events.

Confidence Score: 5/5

This looks safe to merge.

• No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
aw_transform/merge_subwatcher_fields.py	Updates overlap collection and segment matching so zero-duration bases are preserved while zero-duration sub events do not color durationful bases.
tests/test_transforms.py	Adds focused tests for the new zero-duration base and sub-event behavior.

_{Reviews (1): Last reviewed commit: "fix(transform): handle zero-duration bas..." | Re-trigger Greptile}