torch.iinfo() support for sub byte dtypes

[vayuda]

Add an override for torch.iinfo for sub byte dtypes so that the following doesn't happen:

[supriyar]

@jerryzh168 should this be moved to pytorch/pytorch?

[jerryzh168]

this should be be implemented in tensor subclass I think, will need to figure out what torch.iinfo calls in the end, I'll take a look next week