[ShawnChen-Sirius]

Summary

• Fix the macOS process-exit hang exposed by the chdb-core v26.9.0 release check. The addon built and basic tests passed, but a spawned codegen CLI could finish its query and then block inside libchdb static teardown.
• Close live connections and call chdb_shutdown() from the process-wide exit backstop. Register that callback after the first successful chdb_connect() so it runs before the engine teardown; keep Node environment cleanup connection-only.
• Add a bounded async-query process-exit regression and a timeout for the existing codegen subprocess test.

Verification

Tested with the official chdb-core v26.9.0 macOS arm64 release library. The codegen subprocess hung before the fix and exits successfully after it.

• npm test: 24 passed
• npm run test:durable:node: 24 passed
• Vitest excluding agents-tool.test.ts: 461 passed, 11 skipped (42 files)

The excluded network-blackhole integration test timed out in this local environment; the remaining platforms and full suite need PR CI. This does not change the currently pinned engine v26.7.3.

Fixes #97

Note

Shut down chdb-core before Node process exit

• Adds shutdown_on_process_exit in chdb_node.cpp, a process-wide exit handler that closes all registered connections and then runs engine shutdown.
• Moves exit-handler registration in get_default_conn and acquire_session_conn to after a connection opens successfully, so failed connection creation no longer registers cleanup.
• Adds a teardown test in engine-exit.test.ts that runs an async query in a child Node process and asserts it exits within 15 seconds with no stderr output.
• Also pins the MinIO-compatible Silo image (with digest) in the durable-S3 CI job and adds a 15-second timeout to the built CLI end-to-end test.

^{Macroscope summarized fa0afaf.}