Implemented shared S3 audio caching for prepare and restore --include-audio

docs/internal/workspace.md

@@ -32,6 +32,8 @@ Does not own:
 - `pipeline.workspace.cleanup_after_archive`
 - `pipeline.spool.root`
 - `pipeline.spool.delete_audio_after_archive`
+- `pipeline.cache.root`
+- `pipeline.cache.s3_audio`
 - `session.campaign`
 - `session.session_id`
 
@@ -43,7 +45,8 @@ None directly in this subsystem. Stages may use object storage adapters and then
 - Invocation history is persisted per run in run manifests under `runs/{run_id}/manifest.json`.
 - During each run, stage outputs are often written run-local first (`runs/{run_id}/{stage}/outputs/...`) and promoted to canonical session paths after stage success.
 - `manifest.Artifacts` entries record `ProducerRunID` for durable outputs.
-- For S3 audio sessions, `prepare` records spool/work paths and S3 provenance in `manifest.Inputs`.
+- For S3 audio sessions, `prepare` records work/cache paths, S3 provenance, and spool path when the invocation downloaded the object.
+- Durable cache state under `pipeline.cache.root` is not workspace state and is not part of session cleanup semantics.
 
 ## Skip and Resume Behavior
 - Skip/resume decisions are made in `internal/app` (`run_control.go`, `resume.go`) using stage status in the session manifest.