fix(runtime): tighten shutdown and failure events

[jaredjakacky]

Summary

Tightens two v1 lifecycle semantics in the runtime.

• Stops Runtime.Shutdown before StopAll when the caller context has already expired
• Prevents shutdown timeout from marking otherwise healthy workers failed
• Keeps servekitservice fallback cleanup working with a fresh stop context
• Suppresses duplicate non-command FailureEvents when a worker is already failed
• Preserves per-attempt command failure event behavior

Why

Runtime.Shutdown previously used one context across drain, idle wait, and stop. If the context expired during drain or idle wait, shutdown could still call StopAll with an already-expired context. Context-aware workers could then return DeadlineExceeded, causing Workerkit to mark them failed during shutdown.

Separately, if WorkerRuntime.ReportFailure was called during Start and Worker.Start later returned an error, Workerkit could emit two lifecycle failure events for one startup failure path.

Both behaviors make lifecycle and observability semantics noisier than they should be for v1.

Verification

• make test
• make test-race