Change _setindex implementation for tuples to improve constant propagation

[OlivierHnt]

Resolves #39944

[KristofferC]

Would be good to add a test.

[OlivierHnt]

Thank you for the feedback. I added an @inferred test which fails without this PR.

[OlivierHnt]

The failure of /buildbot/tester_linux32 seems unrelated to the PR.
@KristofferC is there anything else I can do to help the reviewing process?

[OlivierHnt]

I would like to "revive" this PR; @KristofferC can you provide some feedback on this?

Let me give some context again: the current _setindex method relies on an arithmetic operation which prevents constant propagation for tuples with (strictly) more than 2 elements. The problem showed up in selectdim for arrays with dimension (strictly) greater than 2.

@mbauman perhaps you could review this PR since it is a faithful follow-up of your suggestion on Discourse (cf. https://discourse.julialang.org/t/selectdim-behaviour-for-arrays-of-dimension-greater-than-2/52211)?

Best regards

Note: the failure of /buildbot/package_win32 and /buildbot/package_win64 appear unrelated.

[KristofferC]

Could you try rebase this on the master branch to perhaps fix the windows CI?

When CI is green, I think this can be merged. Feel free to keep pinging if it gets forgotten again.

[OlivierHnt]

@KristofferC thank you for the feedback. I updated the PR in accordance with your review. All tests have passed.

[OlivierHnt]

Friendly ping about the prospective merging of this PR with the master branch as suggested by @KristofferC 😊