Fix baddbmm handling of "beta" special-case

[Birch-san]

baddbmm() has a beta coefficient, by which bias may be multiplied.

in the special-case where beta=0: bias may be ignored (confirmed by docs):

If beta is 0, then input will be ignored, and nan and inf in it will not be propagated.

we use this in diffusers / stable-diffusion to avoid adding attention bias when none is specified:
https://github.com/huggingface/diffusers/blob/bbab8553224d12f7fd58b0e65b0daf899769ef0b/src/diffusers/models/cross_attention.py#L237

currently, the special-case is determined by a comparison with 1.0 rather than 0.0. it looks like it was copied from alpha's special-case:

coremltools/coremltools/converters/mil/frontend/torch/ops.py

Lines 5274 to 5282 in fa190de

	if beta.val != 1.0:
	# Apply scaling factor beta to the bias.
	bias = mb.mul(x=beta, y=bias, name=bias.name + "_scaled")
	context.add(bias)
	if alpha.val != 1.0:
	# Apply scaling factor alpha to the input.
	batch1 = mb.mul(x=alpha, y=batch1, name=batch1.name + "_scaled")
	context.add(batch1)

changing this, fixed compilation of diffusers' UNetCondition2D model for me (which employs that baddbmm in CrossAttnProcessor).

[aseemw]

Thanks for the PR, can you please add a unit test as well in the torch unit test file, which fails without this change, but passes with it.