gh-125063: Emit slices as constants in the bytecode compiler

[mdboom]

This emits slices as constants in the compiler (when the slices only contain constant members).

I measured a 10% speedup on the cavity_flow benchmark in npbench, which not surprisingly uses a lot of slices, but this kind of code is fairly idiomatic for Numpy. (EDIT: The benchmark as a standalone script)

This does measure as are 1% slower on pyperformance, but that seems largely due to unpack_sequence. I'm going to rerun on more platforms, and also collect stats, since I would expect a measurable reduction in instructions executed.

• Issue: Emit slices as constants in the bytecode compiler #125063

[mdboom]

Here are the largest stat changes:

Name	Before	After	Change
BINARY_SLICE	1,144,280,918	237,406,036	-79.3%
BINARY_SUBSCR	2,095,449,827	3,002,218,240	43.3%
STORE_SLICE	162,427,163	149,989,760	-7.7%
LOAD_CONST	18,743,983,320	17,819,791,524	-4.9%
EXTENDED_ARG	481,860,317	466,468,380	-3.2%
STORE_SUBSCR	876,092,109	888,534,141	1.4%
BUILD_SLICE	211,754,494	209,634,774	-1.0%