New `prod(::AbstractArray{BigInt})` implementation (#59456)
Instead of taking of using a pre-allocated linear product tree (i.e.
`foldl(mul!, arr)`), use a balanced binary product tree with
pre-allocated linear leaves of length 16. This is far from optimal but
already often an order of magnitude or two faster than the current
implementation. It can be sub-optimal by up to a factor of `log2(n)`.
This is reached in the case where there is one big element and the rest
are small and nonzero in which case this implementation performs
`log2(n)` large multiplications while the optimal only performs `1`. The
current implementation is sub-optimal by a factor of `n/log2(n)` in
cases where all the elements are similar sizes.
Fixes #59052
