[OskarEichler]

Problem

When any shard rejects during FuseWorker initialization, the rejected promise remains cached and the partially initialized workers remain alive. Every later search()/add() call immediately receives that same rejection, even if a fresh worker generation would initialize successfully. Pending initialization calls can also retain their handlers.

Fix

• Clear the cached initialization promise after that exact attempt fails.
• Terminate every shard from the failed generation.
• Reject and clear the remaining pending initialization calls with the same failure.
• Reuse the cleanup path for explicit terminate().
• Add a regression in which the first generation fails, is fully terminated, and the next search creates a fresh successful generation.
• Rebuild the committed worker proxy and declarations.

This changes only initialization-failure recovery; runtime worker replacement remains outside this PR.

Verification

• Focused worker suite: 19 tests passed.
• Full suite: 22 files and 450 tests passed.
• ESLint and TypeScript checks passed.
• Worker proxy distributions and declarations rebuilt successfully.
• Changed files pass Prettier and git diff --check.