fix(core): preserve progressing sessions during location cleanup - #51583

Issue for this PR

Fixes #51343 (progressing Sessions and owned shell/subagent work). Related: #48691, #44471.

Type of change

• Bug fix

What does this PR do?

Release inactive Sessions without interrupting other progressing Sessions or killing their foreground/background work.

Special things to note

• The timeout remains one hour; unrelated Sessions and viewing do not extend an unanswered question or permission.
• Shell/subagent ownership uses existing process-local lifetimes, not durable running markers; children retain their own stall timeout.
• No timeout configuration, PTY lifetime, or reconnect-modal changes are included.

Change outline

 inactivity sweep
- directory deadline expires → interrupt every session there
+ Session progress or human input → renew only its Session deadline
+ live shell → retain its Session and original Location through cleanup
+ live foreground/background subagent → retain its parent
+ last owned work cleans up → start a fresh one-hour idle window
+ expired Session → interrupt only its execution and cancel its requests
+ idle Location → evict only when no execution/process still needs it

How did you verify your code works?

• Ran the nine related core test files: 162 passed, 19 skipped; six shell-syntax failures also reproduce on the unchanged synced baseline with macOS /bin/bash 3.2.
• Replaced the inactivity cases and extended an existing Session-shell test. Real execution/forms/SQLite and a file-gated process cover isolation, owned work, child stalls, and fresh idle windows; the shell/child regressions fail with ownership hooks removed. Timing uses virtual advances, not an hour-long soak. Browser verification is in-process RPC, not a physical browser or live provider.
• Root bun run check passed: 36/36 typecheck tasks, lint warnings but no errors. git diff --check passed.

Screenshots / recordings

Earlier isolated TUI demonstration: simulated streams drive 3 same-directory Sessions; C's unanswered question expires while A and B continue. The inactivity timeout was temporarily 5 seconds with a 250 ms sweep; neither the override nor simulation harness is included in this PR.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR