[uv] use torch cpu integration instead of env var

[rmehri01]

Why

• [uv] use torch cpu instead of cuda by default #320 doesn't seem to work on replit, I think we're running into uv sync: Distribution markupsafe==3.0.2 @ registry+https://download.pytorch.org/whl/cpu can't be installed because it doesn't have a source distribution or wheel for the current platform astral-sh/uv#9647
• Using the pytorch integration
  with the latest nightly uv seems to fix the issue

What changed

• Check to see if torch is in the added packages or already in the spec file and if so, idempotently add the torch cpu override to pyproject.toml using toml-editor

Test plan

• Running locally as well as on replit should work now:

[blast-hardcheese]

Nearly there! You should also be able to test this in a repl, now that the latest toml-editor has rolled out 💪

[blast-hardcheese]

Works great!