fix: continue draining queue after postMessage errors

[SyMind]

Summary

• report whether WorkerInfo.postTask() successfully posted a task
• keep draining the pending queue when postMessage() synchronously rejects an unserializable task
• add a regression test with a blocked worker, an unserializable queued task, and a valid task behind it

Problem

When a worker becomes available, _onWorkerAvailable() removes one task from the queue and returns after calling postTask(). If postMessage() throws, postTask() rejects that task but does not occupy the worker. The scheduler still returns, leaving the worker idle and later queued tasks stuck indefinitely.

This surfaced while removing a redundant structured-clone preflight from Rspack's parallel loader path. A non-cloneable loader input correctly rejected, but the shared pool stopped scheduling later builds.

The fix keeps the task-level rejection and continues the existing scheduling loop only when posting failed. Successful task dispatch remains unchanged and does not add a preflight clone.

Validation

• vitest run — 99 tests passed
• tsc --noEmit
• eslint --max-warnings=0
• tsdown