gh-116976: Merge _CHECK_STACK_SPACE into preceding _CHECK_FUNCTION_EXACT_ARGS

[gvanrossum]

This reduces the combined JIT template size by about 75 bytes (on Intel; 88 bytes on my M3, where everything is bigger, even the battery :-).

These opcodes always occur in a pair, and _CHECK_STACK_SPACE does some duplicate work (code = func->func_code). The latter doesn't use any cache entries. Both contain 2-3 DEOPT_IF() calls.

I'll run a benchmark once the tests pass.

• Issue: Combine _CHECK_FUNCTION_EXACT_ARGS and _CHECK_STACK_SPACE uops? #116976

[brandtbucher]

Was the idea behind splitting these that we could maybe sometimes optimize away one, but not both?

For example, if we already checked a larger stack size, we could remove this one. Similarly, if we have already guarded against this function’s version, we could probably avoid doing it again.

If that’s the case, then maybe they make sense to leave apart. Otherwise, the change looks good.

[gvanrossum]

Still waiting for @markshannon.

Benchmarks look neutral:
https://github.com/faster-cpython/benchmarking-public/tree/main/results/bm-20240318-3.13.0a5+-4435189-JIT

[gvanrossum]

Mark prefers to keep them separate. We should do a PR to actually eliminate this guard.