Improve performance for switch in ceval.c when using MSVC #91719
Description
We've received reports of MSVC not generating optimal code, e.g. gh-89279.
One possible improvement would be to get the big switch statement in ceval.c to generate better code. It's been rumored that MSVC will generate essentially a computed goto if all cases are filled.
See my investigations at faster-cpython/ideas#321 (comment)