This fixes issue #5274.
For integer types, any string-based type-checking is replaced with a list-based type-checking using the type lists provided in `theano.tensor` (int_dtypes, uint_dtypes, integer_dtypes, etc.).

NB:
* Module `theano.sparse` have its own type lists, so they are used in this module instead of `theano.tensor` lists.
* I have tried to check the types with `discrete_dtypes` where it seems relevant (e.g. in `grad()` functions), but further analyzes may be required. That is the type list to use to fix issue #5274.