Should we make int32's configurable?

Just noticing that we rely a lot on 32-bit integers in datashader quite a bit - grepping for 'i4' and 'int32'. Should we make these configurable at some point? Either way, it would be nice if we had more tests that exercise inputs that exceed 2.15B, and verify that we aren't encountering subtle overflow issues (probably easier said than done). What do you think?

Start by grepping the codebase for the 'i4' and 'int32' usages mentioned in the issue, then identify which paths depend on 32-bit integers. Review how inputs exceeding 2.15B would be exercised and define tests that detect overflow; done requires a settled direction on configurability and coverage for those inputs.

Written by the indexing model from the issue text.