AIC model caching

[gkmngrgn]

Summary

• AIC model caching: Read-only AIC models are now shared across AICFilter instances via a singleton AICModelManager in aic_filter.py. Multiple filters using the same model path or (model_id, model_download_dir) share one loaded Model and avoid duplicate loads.
• Reference counting: The manager acquires a reference when a filter calls acquire() in start() and releases it in stop(). When the last reference is dropped, the model is evicted from the cache.
• Concurrent load deduplication: If several filters request the same uncached model at once, only one load runs; the rest await the same in-flight task and then use the cached result.
• Internal helpers: Cache key is built by _get_cache_key(); reference updates use _increment_reference() and _store_new_reference() so logic is centralized and model is never optional where it is used.

Testing

• uv run pytest tests/test_aic_filter.py tests/test_aic_vad.py -v — all 28 tests pass, including:
  • test_model_manager_reference_count: ref count increments on get_model, decrements on release, model evicted at 0.
  • test_model_manager_concurrent_load_deduplication: concurrent get_model calls for the same key trigger a single Model.from_file and all callers receive the same model.

No breaking changes: AICFilter’s public API (constructor and start/stop) is unchanged.

[codecov]

Codecov Report

❌ Patch coverage is 0% with 77 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/pipecat/audio/filters/aic_filter.py	0.00%	77 Missing ⚠️

Files with missing lines	Coverage Δ
src/pipecat/audio/filters/aic_filter.py	0.00% <0.00%> (ø)

... and 86 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[markbackman]

LGTM