MAINT: remove incidental imports from private modules

[h-vetinari]

This has now grown into a full-blown fix for #19905; while we're targetting 1.14 with this, I wanted to check if we're actually warning in all cases (after discovering some potential discrepancies).

---

Previously:

Some more follow-ups from the work for #18279, same as #19893.

I've tried & failed to import these on 1.11 resp 1.12; the imports are fully ignored AFAICT due to the presence of __all__ in those private modules.

Not sure what the intention there was @j-bowhay, but in any case, this should be good to remove IMO.

[mdhaber]

These imports were intentional; they were part of the plan to remove these unnecessary imports (and other pseudo-public attributes) in a backward compatible way. The current behavior is, e.g.:

from scipy.optimize.slsqp import exp

which produces the desired warning message.

DeprecationWarning: `scipy.optimize.slsqp.exp` is deprecated along with the `scipy.optimize.slsqp` namespace. `scipy.optimize.slsqp.exp` will be removed in SciPy 1.13.0, and the `scipy.optimize.slsqp` namespace will be removed in SciPy 2.0.0.

The tests checking this behavior are failing. But I guess they can be removed now too since we said they'd be removed in SciPy 1.13.0.

[h-vetinari]

The current behavior is, e.g.:

In one of these discussions the reason given for keeping these imports was that they once were part of __all__. As it happens, that wasn't the case anymore for the import I removed in #19893, but still affects some of the functions here.

These imports were intentional and part of the plan to fix gh-18279.

In any case, we've scheduled them for removal in 1.13, so the timing is still OK. 🙃

[mdhaber]

So yeah, go ahead and remove the tests. It's fine with me to take all these out now, even if 1.13 is coming earlier than usual.

[h-vetinari]

In any case, we've scheduled them for removal in 1.13, so the timing is still OK. 🙃

Actually, we've delayed this to 1.14 in #19892. So I'll wait until 1.13 is out.

[h-vetinari]

@j-bowhay this is mainly a follow-up to your efforts around removing those incidental imports - I think it's ready and would appreciate your review (I think we were close already a month ago; I suggest to finish this now).

[j-bowhay]

Both this and #19673 are very much still on my radar I just need to finish a few exams and assignments before I can spend more than 10 minutes looking at something! I will get to this in May which I assume is fine with the 1.14 timeline.

[j-bowhay]

Friendly ping @h-vetinari as I am keen to get this in before branching for the next release. Can you resolve the merge conflicts?

[j-bowhay]

It looks like you've accidentally picked up a submodule change

[h-vetinari]

While I generally ask not to, this PR can (or even: should) be squash-merged.

[h-vetinari]

The failure here is unrelated:

 =========================== short test summary info ===========================
FAILED scipy\optimize\tests\test_constraint_conversion.py::TestNewToOld::test_individual_constraint_objects
= 1 failed, 63629 passed, 3042 skipped, 246 xfailed, 14 xpassed in 648.81s (0:10:48) =

[h-vetinari]

Any further comments here @j-bowhay et al.? 🙃

[j-bowhay]

LGTM and thanks for tackling @h-vetinari

[h-vetinari]

Thanks for reviews & merging!