fix: harden worker lifecycle concurrency and cleanup

[jaredjakacky]

Summary

Hardens Workerkit lifecycle behavior against several semantic races involving startup, loop shutdown, command completion, and concurrent lifecycle operations.

Changes

• Preserve asynchronous failures reported during worker startup.
• Scope WorkerRuntime handles and commands to lifecycle generations.
• Prevent stale callbacks from mutating restarted workers.
• Serialize lifecycle operations while keeping command dispatch concurrent.
• Prevent duplicate loops after timed-out stops.
• Ensure loop failures are published before stop completion.
• Coordinate loop cleanup hooks across concurrent and repeated stops.
• Preserve stopped lifecycle when in-flight commands fail or panic late.
• Make observer test helpers safe for concurrent callbacks.
• Clarify direct Stop, graceful shutdown, and observer contracts.
• Fix cleanup being skipped after a LoopWorker restart fails during its start hook.

No new exported APIs, packages, or dependencies are introduced.

Verification

• make verify
• make test-race
• go vet ./...
• Full repository race tests with shuffle across 20 runs
• Core race tests with shuffle across 100 runs
• Focused LoopWorker race tests across 100 runs
• govulncheck reports no vulnerabilities