Optimize element shape computation by replacing numpy.prod with numpy.multiply.reduce

[SaFE-APIOpt]

fiat/FIAT/mixed.py

Line 70 in 637894a

sub_cmps = [0] + list(numpy.prod(e.value_shape(), dtype=int)

In the line：
sub_cmps = [0] + list(numpy.prod(e.value_shape(), dtype=int) for e in self.elements())
it is recommended to replace numpy.prod() with numpy.multiply.reduce() for better performance. While both functions yield the same result, numpy.prod() is a wrapper around numpy.multiply.reduce() and includes additional overhead such as input validation, type conversion, and internal logic for handling parameters like dtype, keepdims, and initial.

In scenarios where only a straightforward element-wise product is needed—as in computing the total number of components from e.value_shape()—using numpy.multiply.reduce() provides a more efficient execution path with fewer internal function calls. A direct replacement like the following preserves functionality while reducing overhead:
sub_cmps = [0] + list(numpy.multiply.reduce(e.value_shape()).item() for e in self.elements())

[pbrubeck]

Thanks for raising this point. Does this replacement improve performace?

This bit of code was taken from UFL, where it appears many more times.

Since we have this pattern in a few places, it might be good to introduce it as a method value_size().

[rckirby]

I can't imagine that the performance difference in the scope of everything being done in FIAT, but lifting the pattern into a function call could improve readability (and doing whatever is best inside that call makes sense).

Feel free to make pull request.