fix: prevent LogStreamer.Flush deadlock when context is cancelled

[wesleyjellis]

Problem

LogStreamer.Run exits on either the quit channel or ctx.Done(). Flush unconditionally sends on the unbuffered quit channel, so when the context has been cancelled (SIGINT/SIGTERM while a job is running) there is no receiver and Flush blocks forever.

Downstream effect in jobWorker: the worker goroutine hangs inside Flush, wg.Wait() in doRun never returns, the runner never unregisters, and the process hangs until SIGKILL (the signal handler only cancels the context on the first signal).

Fix

• Run now closes a done channel on exit.
• Flush selects between sending quit and observing done, then waits on done — replacing the fragile time.Sleep(200ms) handshake with real synchronization (this also removes a latent data race on the ring buffer between Run and Flush's drain).
• The wait loop in Flush breaks out early when Run has already exited instead of spinning until the 30s timeout.
• The final drain now consumes all remaining buffered lines, not just one per stream, since Run may have exited early with complete lines still buffered.

Testing

• Added TestLogStreamerFlushAfterContextCancel, which deadlocks on main and passes with this change.
• go test -race ./pkg/... passes.

🤖 Generated with Claude Code