[vsolano9]

Addresses item 6 of #64: the debug server no longer shares main's shutdown context.

Cause

debugSrv.Shutdown(shutdownCtx) reused the 5s context that srv.Shutdown had just been given. Shutdown waits for in-flight requests, and a pprof CPU profile or trace holds its request open for as many seconds as the caller asked. So one ?seconds=30 profile made every shutdown run that context out. The force-exit watchdog is armed for the same five seconds, so the end of shutdown is a race between main returning and os.Exit(1).

Change

The debug server is stopped with Close(), first, and outside the graceful budget (stopDebugServer in main.go). The issue suggested a context of its own or shutting the debug server down first. Neither stops Shutdown from waiting for the profile to finish, so both would still cost seconds the sessions and plugins need. A profile cut short by a server that is going away is worth nothing. Closing the connection cancels the request context, which ends pprof's wait and releases the CPU profiler.

Item 6 only. Items 5 and 8–10 are untouched.

Tests

• TestStopDebugServerDoesNotWaitForAnInFlightProfile starts the real debug server and opens a 30s CPU profile. It waits until the handler actually holds the CPU profiler (StartCPUProfile fails), with no sleeps. Then it requires stopDebugServer to return in under 1s, the profile request to end, and the profiler to be released.
• TestStopDebugServerAcceptsNil covers the bind = "" case, which main now passes straight through.

Red proof: with stopDebugServer changed back to Shutdown with a 5s context, the first test fails with stopDebugServer took 5.000370844s with a profile in flight; want under 1s.

Stability: go test -race -count=40 -run '^TestStopDebugServer' . and the same with GOMAXPROCS=1 -count=20: all pass.

End to end

I built the server from main and from this branch. Each ran with only [server] port and [debug] bind = "127.0.0.1:16060", with a ?seconds=30 profile in flight, and then got SIGINT. Two runs, shutdown log lines only:

run 1, main:         exit 0, 5.09s
  Shutting down...
  debug HTTP shutdown error: context deadline exceeded
  Server stopped
  Shutdown timed out, forcing exit
run 2, main:         exit 1, 5.05s
  Shutting down...
  Shutdown timed out, forcing exit
this branch, both:   exit 0, 0.47s and 0.00s
  Shutting down...
  Server stopped

So on main the outcome is a coin flip at the 5s mark. In run 2 the watchdog won, so the process exited 1 before it logged anything about the debug server.

Validation

The four CI commands (Linux, Go 1.27.3):

• gofmt -l .: clean
• go build ./...: clean
• go vet ./...: clean
• go test -race ./...: ok in every package with tests (internal/turn has none)

Prepared with AI assistance (Claude Code). I ran the checks above myself and the outputs are quoted as observed.