Make bundled inputs work with quantized zero inputs (#47407)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/47407
Previously, the code for bundling contiguous single-valued tensors (like
torch.zeros) wasn't working for quantized tensors because it was calling
the `torch.tensor` constructor without passing in the quantizer.
Instead, skip the constructor entirely, which makes this use case work
and also simplifies the code. (Originally, I forgot that
`arg.flatten()[0]` would return a tensor, not a scalar.)
Test Plan: Bundled a quantized zero input and saw it run properly.
Reviewed By: dhruvbird
Differential Revision: D24752890
Pulled By: dreiss
fbshipit-source-id: 26bc4873a71dd44660cc0fcb74c227b754e31663
