fix: rethrow non-longjmp exceptions in execProtocolRawSync protocol loop

[Brown-Sage]

Description

Fixes #1058.

When the WASM backend terminates while execProtocolRawSync is processing a message (e.g. exit(1) after hitting EOF during a COPY ... FROM STDIN issued via exec()), the protocol loop spins forever, synchronously, at 100% CPU. The call never settles, so no timer, Promise.race, or AbortSignal can interrupt it — the process has to be killed from outside.

Root cause

After the backend exits, _PostgresMainLoopOnce() throws ExitStatus. That is not the longjmp sentinel, so it was silently swallowed; the dead backend consumes no input, so #readOffset never advances and _pq_buffer_remaining_data() never drains — the while condition is permanently true, producing a tight call → throw → swallow loop.

Changes

• Rethrow anything that is not the longjmp sentinel (ExitStatus, WASM RuntimeError, …) — the backend is gone; looping cannot help
• Guard the finally's own WASM calls (_PostgresSendReadyForQueryIfNecessary / _pgl_pq_flush) so they cannot mask the original error against a dead runtime

Behavior after the fix (verified against the repro from #1058)

• the killing exec() rejects with ExitStatus { status: 1 }
• follow-up queries reject immediately with the same error instead of spinning
• close() resolves and the event loop stays alive throughout

Regression test

New tests/backend-death.test.ts:

1. COPY ... FROM STDIN rejects instead of hanging
2. queries after the backend died reject promptly
3. close() still resolves

Against unfixed code the test process hangs synchronously — it even defeats vitest's own test timeout, so the baseline run has to be killed externally. With this patch all tests pass.

Also adds a changeset for @electric-sql/pglite.

Fixes #1058

[Brown-Sage]

Closing in favor of #1081 which takes a more conservative approach to the same fix (narrower rethrow scope to preserve extension behavior, plus exitCode restoration). Thanks @psh4607 — one note: a RuntimeError thrown when #readOffset === message.length but _pq_buffer_remaining_data() > 0 still bypasses your rethrow condition, so that corner case can retain the spin.