GH-128563: (Re)move some labels, to simplify implementing tailcalling interpeter.

[markshannon]

Following on from an earlier discussion: It complicates the code a lot if we need to be able to jump from generated code back into manually written labels and vice-versa.

This PR refactors the code so that there are only two transfers from the the code in ceval.c to bytecodes.c and none the other way,
for tier 1 (we can do much the same for tier 2 later).

The two transfers from PyEval_EvalDefault into the interpreter "loop" are:

• goto error; on line 847
• goto start_frame; on line 856

• Issue: A new tail-calling interpreter for significantly better interpreter performance #128563

[Fidget-Spinner]

This should be fine. Can you please run benchmarks to make sure the added jumps don't affect perf? I don't think they should (they are jumps that should be always predicted), but I just want to be safe.

[markshannon]

Any change in performance will just be noise or error.
If anything, there are fewer jumps as all the old jumps to resume_with_error were followed by a jump to error.