fix(test): drain orphaned lmdb cursor wrappers after Read Txn Expiration

[ldt1996]

The Node v24 Unit Test job has been intermittently segfaulting (~50% hit rate) in the LMDB-mode resources test pass, blocking PRs #1351 and #1352. A gdb backtrace from a core dump showed the crash is inside lmdb's mdb_cursor_close at process exit, called by V8's exit-time finalizer pass — lmdb-js's own source warns about this in read.js:706: "this must be closed before the transaction is aborted or it can cause a segmentation fault."

The trigger is the Read Txn Expiration suite in unitTests/resources/txn-tracking.test.js, which is LMDB-only (skips under RocksDB at the before-hook) and deliberately calls checkReadTxnTimeouts() to force-end in-flight read transactions. On Node v24 the exit-time finalizer order then runs mdb_cursor_close on a cursor wrapper whose txn was already aborted, hitting freed native memory — SIGSEGV.

This adds an explicit drain + global.gc() in the suite's after() block so orphaned cursor wrappers are reaped at the end of the suite, while the env is still healthy, rather than at process exit. Adds --expose-gc to mocharc so the GC call is available.

Verified by running the v24 Unit Test job 4 times in a row on this branch (1 initial + 3 reruns): 4/4 pass. Against the prior ~50% per-run failure rate the probability of 4 consecutive passes by luck is ~6%, so the fix is doing something.

This is a targeted mitigation; the underlying lmdb-js cursor-vs-txn lifecycle hazard remains and should be reported upstream.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[claude]

Reviewed; no blockers found.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[kriszyp]

Clean fix for Node v24 LMDB cursor finalizer segfault. LGTM! 🟢

Reviewed by Claude Sonnet 4.6