Fix buffer size computation (#660)

* Use numpy.prod (keep data type) for computing buffer size

* Add comment on computation of 'size'

* Check individual shape components

* Force computation on uint type

* Add check on each dimension datatype

* Add test_allocation_with_various_shape_scalar_types

* Fix flake8 issues

* Fix shape check

* Replace any() with np.any()

* Shape comp: comment phrasing, convert to array only once

Co-authored-by: Andreas Kloeckner <inform@tiker.net>