improve performance and add benchmarks for large libraries (≈2+ hours footage / many clips)

[dvschultz]

Summary

Profile and optimize performance limits around ~2 hours of footage/many clips; add a repeatable benchmark/regression harness to prevent future slowdowns and unbounded memory growth.

Source

• File: Multiple sources
• Path: Multiple tutorial transcript sources

Evidence

App struggles at ~2 hours of footage; observed slowness with lots of clips and performance rebuilds (GMT20260505-000858_Recording.m4a 124–142, 4345–4381)

Acceptance Criteria

• With a dataset of at least ~2 hours of footage (or synthetic project with thousands of clips), app remains responsive while browsing Analyze/Sequence views (no prolonged UI freezes)
• Add a documented benchmark script/procedure and record baseline metrics (load time, list render/scroll latency, memory usage)
• No unbounded memory growth when loading clip lists/thumbnails (verified via profiling)

Implementation Notes

Move heavy work off UI thread; implement pagination/virtualized views; ensure DB indices; consider compact storage for embeddings/metadata.

Extraction confidence: 0.76