- fnode capi shim: single VM+mutex -> VM pool (FIVENODE_VM_POOL, default 4;
1=serial, N=parallel). Each request checks out its own VM so PRG runs
concurrently across libuv worker threads.
- per-request data keyed by VM (FN_NAPI_REQ via ctx.T.VM()) -- no shared
capiReq race.
- napibridge: per-VM handle tracking; ReleaseAll(vm) auto-ends only that
request's npm handles (parallel-safe auto-__end__). FN_AWAIT replaces the
reserved Five AWAIT keyword (Clipper-compat, no gengo codegen -> NIL).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
6a8ada16b2