fix: health endpoint now checks KataGo process status

[stubbi]

Summary

• Health endpoint returns 503 when KataGo process is not alive (was always returning 200)
• Fixed query_version requests to include required id field
• Fixes keepalive ping error spam in KataGo logs every 30 seconds

Problem

The health endpoint was always returning 200 OK with "status": "healthy" even when the KataGo process had crashed. This prevented Salad Cloud's liveness probes from detecting failures and restarting the container.

Additionally, the keepalive ping was sending malformed requests without the required id field, causing error log spam:

Error: {"error":"Request must have a string \"id\" field"}

Solution

1. Health endpoint now checks engine.is_alive() and returns 503 Service Unavailable when KataGo is not running
2. Added id field to keepalive ping and query_version requests per KataGo protocol requirements

Test plan

• Verify health endpoint returns 200 when KataGo is running
• Verify health endpoint returns 503 when KataGo process dies
• Verify no more "Request must have a string id field" errors in logs

🤖 Generated with Claude Code