Lua stack depth outside the Prolog engine: shen.curry, post-reader walkers, user non-tail recursion (~12k frames) #93

This follows up #85 and #88. #90 removes the native Prolog engine's own depth limit by running deep proofs on fresh coroutine stacks, and #91 makes the reader iterative. Several capcov repros still overflow LuaJIT's fixed ~65500-slot Lua stack. The overflow now happens outside the Prolog engine and the reader, in non-tail-recursive Lua code: kernel KL tree walkers and user code. shen-go 1f587de succeeds on all of them, because it has a growable stack.

Results with #89, #90 and #91 merged together:

All of these hit the same limit: each Lua frame takes stack slots and the stack can't grow past ~65500 slots. Possible directions:

• Kernel walkers (shen.curry, shen.find-types, shen.process-applications, macroexpansion, ...): native iterative overrides, as was done for the comment scanners and Reader fails on list literals over ~1550 elements with a misleading reader error near here: (suspected Lua stack overflow) #88.
• The general case, user code included: run evaluation on a fresh coroutine when the stack is deep, as fix(prolog): run deep proofs on fresh coroutine stacks (#85) #90 does for the Prolog engine. Every coroutine has its own ~65500-slot stack. The hard part is finding a cheap trigger for compiled KL, which has no single thaw point to count at.

Generated by Claude Code