Upstream the mruby stack_init() nil-init fix (multi-VM-per-process crash) #60

PR #59 fixed a latent upstream mruby bug (present through mruby 3.4.0): stack_init() in vendor/mruby/src/vm.c mallocs the initial 128-slot VM stack but never nil-initializes it, unlike stack_extend_alloc() which stack_clear()s its grown region. GC's mark_context_stack() then reads uninitialized slots — harmless on a first boot (fresh zero mmap) but on a later mrb_open (after prior VMs dirtied the heap), a leftover word is marked as a bogus object pointer and dereferenced → SEGV. Linux-only (glibc/musl recycle dirty memory eagerly; macOS libmalloc masked it). Valgrind-confirmed dirty-memory-reuse, not OOB.

Our fix (one line, stack_clear(c->stbase, STACK_INIT_SIZE) in stack_init) is a local vendored patch documented in vendor/mruby/README.md as re-apply-on-bump. It should be reported/PR'd upstream to mruby — it affects anyone embedding multiple VMs per process. Task: file the upstream mruby issue + PR, link here; once upstream merges + we bump past it, drop the local patch.

Refs #56, PR #59. https://claude.ai/code/session_011szWvquoss1yNX7KWSKCaM