segfault in harr_visit()

[llasram]

I found another one!

$ clang --version | head -n1
Debian clang version 19.1.4 (1)

$ git rev-parse --short HEAD
5ecd1858

$ make o3n
0/0 in 0.1s

$ ./BQN -e '(0⋈⊢)•_while_(1) ⟨⟩'
Segmentation fault

[dzaima]

This is essentially intended currently; problem is that the object freeing (and GC visiting before that) is recursive, and thus deeply-nested objects can exhaust the stack. And while it could be fixed, it'd be at the cost of some overhead on every object free.