tp: add INTERVAL FLATTEN to pipelines

[LalitMaganti]

FROM spans
|> INTERVAL FLATTEN PER cpu AGGREGATE COUNT(*) AS n, SUM(weight) AS w

Cuts overlapping rows at every start and end into disjoint segments, one
row per segment with its bounds, the PER keys and the aggregates. A row of
no width becomes a segment of no width counting the rows spanning it.
Bounds follow INTERVAL INTERSECTION: null ts or dur is skipped, negative
is an error. Only COUNT(*) and SUM for now.

The operator needs its input grouped by the keys and ordered by ts within
each group. Lowering tracks what order is known and adds Sort and GroupBy
only when needed, e.g. no Sort for a dataframe sorted on ts. Its output
keeps that order, so a following stage needs neither.

Against the stdlib macros it replaces, on real traces (release build,
materializing the result as a table):

Workload	Macro	Macro time	FLATTEN time
sched by machine_id	intervals_overlap_count_by_group!	329 ms	35 ms
thread_state by state	intervals_overlap_count_by_group!	986 ms	94 ms
slice by track_id	intervals_overlap_count_by_group!	1213 ms	62 ms
sched	intervals_overlap_count!	264 ms	28 ms
Chrome slices	interval_self_intersect!	1352 ms	7 ms

[github-actions]

🎨 Perfetto UI Builds & Tests

• 9ed001a (2026-09-30 00:06 UTC): https://storage.googleapis.com/perfetto-ci-artifacts/gh-36647563404-1-ui/ui/index.html