Host-side memory handling gaps found while probing #248 (malloc NULL, getOwnPropertyNames free, HostRefMap, asyncify unwind buffer) #274

Stack: #248 → #276 → #277 → (#278 | #279 → #280 → #281 → #282). Each PR body has a link to review just its own diff.

While stress-testing #248 (heap growth) I found four related problems. All reproduce identically with and without #248. Repros were run against the published 0.32.0 wasm variants with a locally built quickjs-emscripten-core; I could not build wasm locally, so C-side items are untested proposals.

1. Unchecked _malloc in ModuleMemory (toPointerArray, newTypedArray, newHeapCharPointer, newHeapBufferPointer). When the 2 GiB heap is full _malloc returns 0 and the host then writes at address 0, corrupting the module (later calls fail with "memory access out of bounds"). Proposal: throw. (Done in the PR below.) The C side has the same pattern (fixed in Handle allocation failure in newContext, heapValueHandle, getOwnPropertyNames (part of #275) #280 for QTS_GetOwnPropertyNames and jsvalue_to_heap) (QTS_GetOwnPropertyNames malloc(...) unchecked at c/interface.c ~929; jsvalue_to_heap returns NULL and nothing checks it).
2. getOwnPropertyNames frees a plain-malloc'd array with QTS_FreeVoidPointer (js_free), which decrements QuickJS's malloc_count/malloc_size, so memory accounting (and setMemoryLimit) drifts. Proposal: free with the module's _free. (Done in the PR below.) Also note that array is over-allocated using sizeof(JSValue) (8 on wasm32) for 4-byte pointers (C side, not changed).
3. HostRefMap freelist uses Array#shift: draining N freed ids is O(N²) (11s at 200k ids in a pure-JS repro). Also deleting an already-deleted id in a non-empty group queued it twice, so it was handed out twice. Proposal: head-index queue; throw on double delete. (Done in the PR below.)
4. Asyncify unwind buffer is fixed at 80 KB (ASYNCIFY_STACK_SIZE=81920): suspending an async host function at moderate guest recursion depth crashes the module, and a crashed module stays dead. Partly addressed in Size asyncify buffer from max stack size so deep suspensions cannot abort (part of #274) #282: needs a build-flag change (or a clearer error) and a wasm build to verify. Repro: function f(d){return d===0?hostSleep():f(d-1)+0} with an asyncified hostSleep, increasing depth.

PR for items 1-3: #277 (stacked on #276 and #248). Item 4: fixed for explicit setMaxStackSize in #282 (stacked on #281); the default-size question is raised there.