Support alternative pip installers (uv) via get_pip_cmd

[yuxuan-z19]

Currently, installation scripts always invoke pip via sys.executable -m pip in subprocesses.
This PR introduces get_pip_cmd() (same as PR#2659 for torchbench), which:

• Uses the PIP_MODULE environment variable to override the installer if set.

  PIP_MODULE="uv pip" python install.py

• Defaults to sys.executable -m pip otherwise.

[xuzhao9]

Thanks for the contribution. Unfortunately, our build environment is not ready for uv yet, even when uv is available.

The main blocker is that we need conda env clone to construct and hot-switch multiple environments with the default pytorch triton, triton-main, and meta's triton fork. Right now, uv does not support multiple environments for single project.

We can accept the env PIP_MODULE, but we need to use sys.executable -m pip otherwise.

[yuxuan-z19]

Got it @xuzhao9! I’ve removed the uv fallback, so the installer now only honors PIP_MODULE when provided and otherwise falls back to sys.executable -m pip. Could you take a look and let me know if this aligns with what you had in mind?

[xuzhao9]

LGTM