[WIP] Reimplementing np.mean with overload and adding axis parameter

[luk-f-a]

Partially addresses #1269 and #4245
Requires #4472 (first 3 commits come from that PR), #5060 and #5711, which fix intermediate problems.

[luk-f-a]

closing to open a new branch to avoid merge conflicts with #4472

[luk-f-a]

Status: The current mean is able to deal with np.datetime64 but an overload-based version fails because of #4540. It was agreed with @stuartarchibald that breaking support for np.datetime64 would be wrong, so this PR is on hold until that issue is resolved.

[luk-f-a]

current status: #5060 is almost done and will fix #4540 which was blocking this PR.
However, another issue came up: timedelta could not be divided by a number, which is needed to calculate the mean. So I fixed that in #5711. However, that's not the end of it, there's an additional issue: after changing the implementation from lower_builtin to overload the new function is not registered in npydecl.registry. This is a problem for ir_utils.get_np_ufunc_typ, which looks for the function there, leading to a failure in the tests test_parfor.test_var and test_parfor.test_std

[N720720]

It looks like we are so close to get this working. How can we register the new function in npydecl.registry? How to contribute to this issue?

[sklam]

I pushed a merge main commit to resolve conflict so we get new CI run. There are changes to get_np_ufunc_typ that may change things.

[github-actions]

This pull request is marked as stale as it has had no activity in the past 3 months. Please respond to this comment if you're still interested in working on this. Many thanks!