Implement prod_all and prod_keepdim

prod_all / prod_keepdim are like sum_all / sum_keepdim but for product. [numpy](https://numpy.org/doc/stable/reference/generated/numpy.prod.html) has this implemented. It is the torch.prod equivalent with the dim argument. It will be good to have this implemented.

The library I am working on [burn](https://github.com/tracel-ai/burn) - is adding this functionality but I have to implement it for candle in the burn codebase.

Start by locating the existing sum_all and sum_keepdim implementations in candle and compare their behavior with the requested prod_all and prod_keepdim operations. Use numpy.prod and torch.prod with a dim argument as references; done means both product operations support the corresponding dimension behavior and keep-dimension variant.

Written by the indexing model from the issue text.