Occasional memory corruption in the interpreter

[straight-shoota]

The CI job for testing the interpreter occasionally fails with an invalid memory access. Example: https://github.com/crystal-lang/crystal/actions/runs/3128470721/jobs/5076354940

Invalid memory access (signal 11) at address 0x8
  [0x560cf663c2b6] ?? +94613673329334 in /mnt/.build/crystal
  [0x560cf663c27d] ?? +94613673329277 in /mnt/.build/crystal
  [0x7f2347ded420] ?? +139789506368544 in /lib/x86_64-linux-gnu/libpthread.so.0
  [0x560cf6302b97] ?? +94613669948311 in /mnt/.build/crystal
  [0x560cf60c4ec5] ?? +94613667598021 in /mnt/.build/crystal
  [0x560cf670aaaa] ?? +94613674175146 in /mnt/.build/crystal
  [0x7f2347d75e06] ?? +139789505879558 in /lib/x86_64-linux-gnu/libffi.so.7
  [0x7f2347d76188] ?? +139789505880456 in /lib/x86_64-linux-gnu/libffi.so.7
  [0x560cf675c031] GC_inner_start_routine +81 in /mnt/.build/crystal
  [0x560cf6753c63] GC_call_with_stack_base +19 in /mnt/.build/crystal
  [0x7f2347de1609] ?? +139789506319881 in /lib/x86_64-linux-gnu/libpthread.so.0
  [0x7f2347c81133] clone +67 in /lib/x86_64-linux-gnu/libc.so.6
  [0x0] ???

The failure does not reproduce consistently.

I think it first appeared roughly together with #12495 which made it look like different symptom of the same bug. But it was not fixed by #12496 which completely reset the effects of #12456 (which caused problems in the interpreter).
So it's most likely a different problem. But it might be related. The very similar stack trace suggests that (it's not repeating, though).

[asterite]

It would be nice to run interpreter specs with -v to see which spec is failing.

[straight-shoota]

I think I tried that in https://github.com/straight-shoota/crystal/actions/runs/3077977934/jobs/4973221240. But the error shown there should be fixed by #12496. I'll run it again (https://github.com/straight-shoota/crystal/actions/runs/3136585192/jobs/5093685998).