Implement multidimensional fast_callable

[videlec]

We implement arrays of fast_callable.
We use it when there is a need to consider
gradient or hessian of multidimensional functions
(typically numerical/optimize.py).

Such arrays should be compatible with
numpy arrays as with #29235.

Depends on #29235

CC: @slel @orlitzky

Component: numerical

Issue created by migration from https://trac.sagemath.org/ticket/29236

[mkoeppe]

comment:1

Batch modifying tickets that will likely not be ready for 9.1, based on a review of the ticket title, branch/review status, and last modification date.

[mkoeppe]

comment:3

Setting new milestone based on a cursory review of ticket status, priority, and last modification date.

[nbruin]

comment:6

For multi-dimensional expressions it makes a lot of sense to analyze the expressions for common subexpressions and re-use partial results on the way.

[mkoeppe]

comment:7

It would be good to consider this in the larger context of graph representations that modern libraries such as pytorch provide. There's little point in rolling this our own.

[mkoeppe]

comment:8

Also possibly related: https://docs.python.org/3.10/whatsnew/3.8.html#pep-590-vectorcall-a-fast-calling-protocol-for-cpython