fix: serialize close with in-flight queries

[seanwessmith]

What changed

• share one close promise across repeated calls
• wait for previously-started transactions and queries before deactivating the WASM backend
• reject operations started after close begins, including while initialization is still completing
• keep lock coordination separate from the existing shutdown routine
• add bounded regression coverage for query, transaction, initialization, and repeated-close behavior
• add the required patch changeset

Why

A query can pass _checkReady() and still be waiting between protocol phases when close() deactivates the backend. The query then enters execProtocolRawSync() against an inactive backend and can block the event loop permanently.

The close path now preserves call order: work invoked before close() drains under the existing transaction/query mutex order, while work invoked afterward sees PGlite is closing.

Fixes #1084.

Related work

#1063 addresses a separate close race with detached filesystem syncs. Both changes affect shutdown ordering and may require conflict resolution if #1063 lands first; this PR specifically serializes close() with in-flight query and transaction work.

Validation

Using the current-main Node 24 WASM artifact from #1087, from packages/pglite:

../../node_modules/.bin/tsup
../../node_modules/.bin/tsc --noEmit
../../node_modules/.bin/eslint ./src ./tests --report-unused-disable-directives --max-warnings 0
../../node_modules/.bin/prettier --check ./src ./tests
./node_modules/.bin/vitest run tests/targets/runtimes/node-close.test.js
./node_modules/.bin/vitest run tests/targets/runtimes/node-*.test.js

• focused close regressions: 3 passed
• Node runtime suite: 13 passed