gh-100817: Speed up copy.deepcopy calls on slice objects

[sobolevn]

See the original issue for microbenchmarks.

• Issue: Make copy.deepcopy faster for slice objects #100817

[AlexWaygood]

This PR results in a behaviour change for the following edge case.

On main:

>>> s = slice([1, 2], [3, 4], [5, 6])
>>> from copy import deepcopy
>>> t = deepcopy(s)
>>> t.start is s.start
False

With this PR:

>>> s = slice([1, 2], [3, 4], [5, 6])
>>> from copy import deepcopy
>>> t = deepcopy(s)
>>> t.start is s.start
True

[sobolevn]

Indeed, I haven't thought about this case. One of my main assumptions was that slice cannot contain nested structures. Because I've only seen int samples in real life. But, if lists are valid and used - this should not be merged. Because deepcopy does not work as promised: it is a shallow copy now.

[AlexWaygood]

Yeah, it's a really obscure edge case, but I can find some examples in the wild: https://grep.app/search?q=%20slice%28%5B&case=true&words=true&filter[lang][0]=Python.

Would you maybe like to add a test for this edge case? I very nearly merged the PR on the basis that all the tests passed, and it seemed like an impressive optimisation :)