Command timeout does not bound total duration while output continues #113

Reproduction

Fresh public agent-sandbox-client[all]==0.4.0a1, Express base sandbox. Run a Python process that prints and flushes every 0.3 seconds for 20 iterations with await sandbox.commands.run(command, timeout=1).

Expected: typed CommandTimeoutError after approximately the execution limit, with the owned process stopped. Actual: successful exit after 6.46 seconds, all 20 output ticks, no timeout. The test sandbox was cleaned up.

Cause

The provider process adapter converts HTTP read inactivity timeouts but does not enforce a total execution deadline. Periodic output resets HTTP inactivity waiting.

Acceptance

• Enforce the existing documented command execution deadline with and without ongoing output.
• Retain typed errors, close the stream and attempt to stop only the owned timed-out PID.
• Add regressions for continuing output, silence, successful short commands and cleanup.
• Reproduce on the real backend, then publish the minimal alpha patch and verify the public installation.

Discovered during the documentation/example rerun; related #111. No credentials or resource identities are included.