Fix implementation of float64 in torch_utils

[viking1304]

Description

Implementation of float64 in torch_utils added in #15815 does not work as intended.

I was getting errors while testing PLMS, DDIM, DDIM CFG++ on my M3 Pro Mac.

TypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead.

This is fixed implementation of float64, which now works as it was intended.

Screenshots/videos:

Checklist:

• I have read contributing wiki page
• I have performed a self-review of my own code
• My code follows the style guidelines
• My code passes tests

[viking1304]

@w-e-w should I close this PR in favor of #16088?

[w-e-w]

TBH, I am not sure

one part of me thinks the new PR is totally reasonable providing 3.9 is nice

but another part of me thinks that we should drop 3.9 compatibility

• we said on the readme we wrote python 3.10
• Google Collab also has upgraded 3.10+
• beginner users should just use the distribution package or use any of those third party installers
• some extension uses match case syntax
• also some and forcing anyone that is 3.9 to upgrade might reduce some version 3.9 completely issues so 3.9 compatibility like this in the future
  not only that I kind of hope that we could "officially" move to 3.11 for the improvements
  love the PEP 657: Fine-grained error location in traceback, that would make reading bug report much easier

I believe some devs are already using 3.11 > > I believe torch support is still not ready in 3.12

TLDR, as I am not the one making the decisions I don't need to use my brain to
I'll just leave the PR open and AUTO decide

[viking1304]

I didn't notice any problem with 3.11 on Mac, but I still use 3.10 for my main installation and when I am testing code before submitting PRs. I also managed to use 3.12 with just a few minor tweaks and issues.

[w-e-w]

https://github.com/pytorch/pytorch/releases/tag/v2.3.0

PyTorch support for Python 3.12 in general is considered experimental. Please use Python version between 3.8 and 3.11 instead. This is an existing issue since PyTorch 2.2.

[AUTOMATIC1111]

Already merged the other one.