Platform-agnostic wheel fixes

[jmanning-stackav]

Description

This PR changes how dependencies for different platforms are specified for conch. By default, a pip install conch-triton-kernels without any extras specified will not install either torch or triton (instead defaulting to use whatever installation is already present in the environment). To install torch/triton for your platform, you can specify your accelerator type via extra, e.g. pip install "conch-triton-kernels[cuda]".

Testing

Please select all that apply.

• Existing unit tests
• Unit tests added by this PR
• Other (please explain)
• This PR is not tested

Test instructions

# Remove old python environment to start from scratch
rm -rf ./.direnv
cd ..
cd conch
# This will create a new virtual env

Nvidia

pip install -e ".[dev, cuda]"

AMD

pip install -e ".[dev, rocm]" --extra-index-url https://download.pytorch.org/whl/rocm6.2.4

pytest

Platforms

Please select all hardware platforms that this PR was tested on.

• Nvidia GPU
• AMD GPU
• Other (please explain)

A10

1409 passed, 27018 skipped in 403.94s (0:06:43)

H100

1583 passed, 26892 skipped in 301.48s (0:05:01)

MI300X

1595 passed, 26880 skipped in 286.78s (0:04:46)