Optimize prod_diagonal default method

[fingolfin]

With some minor modifications it should be possible to turn prod_diagonal(A::ZZMatrix) into the new generic method, replacing the current generic method.

[fieker]

both are sub-optimal: the ZZMatrix function is more GC-friendly, but uses a slow algo for actually computing the product. The generic function copies the diagonal elements into a julia list (GC-bad), but then calls an asymptotically fast version to compute prod....
Ideally the Hecke ProdEnv should be used for large matrices, but I am not sure I have applications where the matrices are large enough and dense so that this matters...
For FqDefault the generics are terrible