torch.ceil, torch.floor should accept a dtype argument

[zasdfgbnm]

🚀 Feature

When a user is doing torch.ceil and torch.floor, it is likely that the user is asking for a long tensor instead of a floating tensor

torch.ceil(x, dtype=torch.long) would make this possible.

cc @mruberry @rgommers

[prajjwal1]

@zasdfgbnm Can I work on this ? If that's okay, could you also provide few hints on tackling this. Thanks!

[zasdfgbnm]

It would be nice if you want to work on this, and I am happy to help. But before doing anything, we need someone on Facebook to confirm if we want to add this feature. Maybe @gchanan ?

[mruberry]

Special case of #36571. We should add a dtype to all unary ufuncs, including ceil and floor.