Optimize blocking calls cuj metrics with scoped materialization and index push-down

The CUJ metrics originally relied on globally scanning all slices via `_android_critical_blocking_calls` and processed `thread_state` virtual tables using unoptimized inequality conditions, causing massive Cartesian nested loops on larger traces.

This CL dynamically accelerates query execution using two components:
1) Restricts materialization scopes cleanly replacing the global dependency with `_android_critical_blocking_calls_cuj`. This limits table mapping dynamically by employing nested CROSS JOIN logic on `relevant_upids`. Crucially, it attaches a structural Perfetto Index on (utid, ts) for instantaneous intersection verifications.
2) Replaces forced inner-loop table evaluations on virtual metrics mapping in `android_blocking_calls_cuj_per_frame_metric` via executing SQLite explicit `CROSS JOIN` conditions to force index-based evaluation on `utid`.

Execution on heavy traces cleanly drops delay times from 127 seconds to just 7 seconds, without sacrificing precision (yields exact output parity).

Test: trace_processor_shell --run-metrics android_blocking_calls_cuj_metric,android_blocking_calls_cuj_per_frame_metric <trace>
Bug: 35267372
TAG=agy
CONV=df6d65a2-a1c7-45c0-8b9c-4e6cafc819a1

Change-Id: Iaeffea6511edbe8b416e76a068ef96bc44289446