MAINT: Refactor differences between cblas_matrixproduct and PyArray_MatrixProduct2

[mattip]

The two functions do very similar things, cblas_matrixproduct is used in both PyArray_MatrixProduct2 (which is the actual implementation of dot) and in array_matmul if ndim <=2 and the dtype is fitting. Otherwise array_matmul goes through einsum, PyArray_MatrixProduct2 has its own continuation. I worked through the logic to compare the two and found points of code duplication:

stage	PyArray_MatrixProduct2	cblas_matrixproduct
detect non-aligned or non-contiguous data	doesn't care	uses PyArray_NewCopy if non-aligned or non-contiguous
create output array	use new_array_for_sum	refactored lines of code to new_array_for_sum in this PR
logic to handle vector-matrix, matrix-matrix, scalar-matrix and visa-versa	uses PyArray_IterAllButAxis	if blocks for each case
chooses function for sum-of-loops	uses out->dtype->vdot	if blocks for each of the level 2 BLAS cases, use out->dtype->vdot for vector/scalar (which will use level 1 BLAS dot functions)

This PR makes the two functions share the output array logic and uses the same semantics for the level 1 BLAS dot functions (scalar-vector)

[mhvk]

Very nice cleanup! Apart from a total nitpick - the unneeded change in numpy/core/src/multiarray/cblasfuncs.h - this looks all OK to me.

[mhvk]

Good to squash before merging, though, that way the commit message is more guaranteed to be what you want.

[mattip]

Squashed, rebased, removed the gratuitous change, and realized that adding a include directive to one file meant I should remove it from the another

[mhvk]

I must admit I do not understand what is the point of possibly passing on result to store the output in when the output is also returned. Since this PR is just about moving stuff so it can be used in multiple places, I think it is better to leave it as is here, but if you have the energy for a clean-up follow-up PR, that might be an idea...

[mattip]

result is the output ndarray to be returned - either out or a freshly allocated ndarray (let's call it retval). retval is the ndarray to be iterated over. If writeback semantics are active then retval is not out rather it is a new ndarray for iterating, and retval->base == out. Then toward the end of the function the writeback semantics are resolved, retval is discared, and the function returns return.

[mhvk]

Thanks for the explanation! If in a future PR you get to touch this file again, maybe add it to the comments...

[mhvk]

I'll merge now. I guess you can copy your table on top for the next clean-up!!