The overall purpose of this PR is to help future work on inter-bytecode optimization and efficient support for tracing events.

This is a bit of an ugly PR as it attempts to do a few things which are conceptually only loosely related, but are interwoven in the code.

1. Specify a set of interpreter "registers", that is the parts of the interpreter state that are worth caching in local variables. This is nothing new, just a bit more explicit. These variables are currently:
   • frame. Pointer to the currentInterpreterFrame
   • names andconsts. Pointers to the names and consts tuples of the current code object. These could be merged into one in future (Merge co_names and co_consts into a single array. faster-cpython/ideas#92)
   • first_instr andnext_instr. Pointers to the first and next instructions.
   • stack_pointer. The evaluation stack pointer.
2. Make all tracing events explicit as macros, and move them into instructions where appropriate.
3. Inline frame exit, entry and resumption where necessary to support above goals.

No meaningfulchange in performance

Seefaster-cpython/ideas#112 for motivation

https://bugs.python.org/issue45753