Ruby launch sessions on Windows end in state 'error' instead of 'stopped' — even for clean, successful scripts

[debugmcpdev]

Found during the #244 default-exception validation sweep (2026-08-04, Windows, dev build at feat/244).

Symptom

A Ruby launch session whose script runs to completion — cleanly or via an unhandled raise — ends with session state error, not stopped. The debuggee itself behaves correctly (all stdout captured, script fully executed); it's the terminal session state that's wrong.

Reproduced with both:

• a clean script (puts twice, exit 0) → state error
• a crashing script (raise "boom") → state error (no exception pause expected or seen — Ruby correctly has no breakOnExceptions default)

Evidence

Log trace (logs/debug-mcp-server-30004.log, both sessions identical shape):

ProxyManager ... Exit: code=1
[SM _updateSessionState] running -> error
"proxy exited during startup"

The rdbg-wrapping proxy process exits with OS code 1 regardless of debuggee outcome, and SessionManager maps that to ERROR. Callers polling for state === 'stopped' as the termination signal never see it for Ruby launches on Windows.

Notes

• Orthogonal to [FEATURE] Consider defaulting breakOnExceptions to 'uncaught' for launch sessions #244 (no exception filters are armed for Ruby; verified zero setExceptionBreakpoints calls and zero warnings in the same log).
• Possibly a regression from the recent adapter-stdio forwarding work (feat: forward adapter-process stdio as debuggee output events — Ruby all platforms, Rust on Windows (#222, #223) #254 forwards rdbg's stdio on all platforms) or a longstanding quick-exit race — the "proxy exited during startup" message suggests the exit beat some launch-phase bookkeeping.
• The /testdebugger sweep of 2026-08-02 reported Ruby green, so if it is a regression it's recent.

Expected

Clean completion → state stopped with the debuggee exit code surfaced (as Python/js do); crash → stopped with non-zero exit code.