ui: optimize BreakdownTracks and AndroidBinderViz performance

[gignat-dev]

Significantly reduce the time it takes to load traces with large breakdown
track hierarchies (like Android Binder) through three optimizations:

1. Projected Path for BreakdownTracks: Instead of running the expensive intervals_overlap_count macro O(N) times (once per track node), pre- calculate all intersections upfront into a shared temporary table. Individual tracks now just run a cheap filter/aggregation against this table. This is enabled for standard count-based breakdown tracks without pivots or complex joins.
2. Batch SQL Table Creation: On the legacy path, CREATE PERFETTO TABLE statements for counter tracks are now accumulated during the hierarchy walk and flushed to the engine in a single batched query, reducing IPC overhead.
3. Parallelize Binder Tracks: The AndroidBinderViz plugin now uses Promise.all to kick off the generation of both server and client track hierarchies concurrently rather than sequentially.

[github-actions]

🎨 Perfetto UI Builds

• b9ad21c (2026-05-28 11:39 UTC): https://storage.googleapis.com/perfetto-ci-artifacts/gh-26572118791-1-ui/ui/index.html