CI:Cygwin: Install packages one at a time to see which one hangs

FIX:CI:Cygwin: Fix quotes in install loop.

CI:Cygwin: Increase timeout on dependency installs

BUG: Include python-including headers first in npstd.hpp

BUG: Include Python-including headers first in fast_loop_macros.h

BUG: Include Python-including headers first in npy_hashtable.cpp

BUG: Import Python.h first in dispatching.cpp

BUG: Include Python.h first in _pocketfft_umath.cpp

https://sourceware.org/pipermail/cygwin/2025-January/257025.html suggests the known problems in Cygwin 3.5.5 might be alleviated by rolling back to Cygwin 3.5.4:

https://sourceware.org/pipermail/cygwin/2025-January/257085.html
suggests this occurs after some step completes.

Running the Python tests suggest the child processes are doing *something*, but attaching gdb to find out what fails: attempts to inspect the stack imply they converted the NULL pointer to a function pointer then called it recursively, which seems unlikely.

It'll take forever, but it might say which tests fail now.

Something's gone wonky with the threading/multiprocessing code.
Not sure how to figure out what, since the debugger shows null backtrace.

Merge in upstream/main

Assumed minor version was a single digit when I wrote this: now checked to be a digit, and allowed to have one or two in the minor version.