python3-scipy: update to 1.12.0.

[tornaria]

Testing the changes

• I tested the changes in this PR: briefly

I enabled tests for scipy let's see if CI can manage or we have to disable in CI (we only run "not slow" tests for -Q).

Updated sagemath since it needs a patch for scipy 1.12 (sagemath/sage#37123). Took the chance to apply a few minor fixes that didn't justify an update on their own.

@ahesford I'll be done with testing in a couple days at most, in principle everything looks good.

BTW: I run scipy tests with export SCIPY_XSLOW=1 and -m xslow which gave me one failure:

FAILED scipy/sparse/linalg/tests/test_onenormest.py::TestOnenormest::test_onenormest_table_6_t_1
====== 1 failed, 321 passed, 40 skipped, 15 xfailed in 9042.78s (2:30:42) ======

Running check normally gave me:

== 45224 passed, 2467 skipped, 149 xfailed, 12 xpassed in 3836.63s (1:03:56) ===

(with -m "not slow") and

== 59100 passed, 2501 skipped, 259 xfailed, 14 xpassed in 5444.99s (1:30:44) ===

[tornaria]

Seems ok, scipy passed check in all arches.

Failure in i686 is one sagemath test, seems transient and completely unrelated to scipy, a timeout in:

...
sage: R.<x> = PolynomialRing(GF(65537), implementation="FLINT") ## line 701 ##
sage: f = R.random_element(9973) * R.random_element(10007) ## line 702 ##
sage: alarm(0.5); f.factor() ## line 703 ##

Most probably the signal got lost, I've seen this in the past. The test looks stupid but I think it's there to make sure f.factor() can be interrupted (say, by hitting ctrl-C, if that signal gets lots one can hit ctrl-C again ofc).

Running the scipy testsuite in CI seems ok, in our three arches with check:

== 45225 passed, 2466 skipped, 147 xfailed, 14 xpassed in 1619.82s (0:26:59) ===
== 44982 passed, 2334 skipped, 550 xfailed, 14 xpassed in 1765.55s (0:29:25) ===
== 45224 passed, 2467 skipped, 147 xfailed, 14 xpassed in 1395.96s (0:23:15) ===

It's weird it takes 26 minutes in CI, 36 minutes in my 8 core desktop, and 1 hour (walltime!) in my 36 core workstation, it seems some tests are scaled by the number of cores.

[tornaria]

I re-pushed because I forgot to git add one patch in sagemath (minor issue that only triggers when sagemath is built with one version of cython and then cython changes version so CI will never see it).

We'll get a chance to see if the i686 failure is indeed transient as I think.

If CI passes this is ok to merge on my side (take your time, no hurry here @ahesford).