Make domain reload detection more deterministic (reduce 20s timeout)

[dsarno]

Summary

Currently, when Unity performs a domain reload (e.g., after running tests or compiling scripts), the MCP connection drops. The server waits up to 20 seconds for Unity to reconnect before failing commands.

This 20-second timeout handles most domain reloads, but it could cause unnecessary delays in agentic workflows where domain reloads happen frequently.

Current Behavior

• UNITY_MCP_SESSION_RESOLVE_MAX_WAIT_S defaults to 20 seconds
• UNITY_MCP_RELOAD_MAX_WAIT_S defaults to 20 seconds
• Server blindly waits up to 20s when no Unity session is available

Proposed Improvement

Make detection more deterministic by:

1. Detecting Unity's actual reload state rather than blindly waiting

   • Check the status file for "reloading" state
   • Monitor for connection attempts during the wait period
   • Use exponential backoff with early exit when connection is established

2. Shorter default with smart detection

   • Default to ~5-10 seconds for normal operations
   • Extend only when domain reload is detected (via status file or heuristics)

3. Different timeouts for different contexts

   • Post-test-suite: longer timeout expected
   • Normal commands: shorter timeout preferred

Related

• Fixed in PR Fix WebSocket connection reliability and domain reload recovery #656 (increased from 2s to 20s to handle domain reloads)
• Issue PropertyConversion failures crash command dispatcher while telemetry continues #654 (dispatcher becomes unreachable after test suite)

Files Involved

• Server/src/transport/plugin_hub.py - UNITY_MCP_SESSION_RESOLVE_MAX_WAIT_S
• Server/src/transport/legacy/unity_connection.py - UNITY_MCP_RELOAD_MAX_WAIT_S

Notes

See TODO comment in unity_connection.py around line 786 for context.