Standardize softmax.py to avoid numpy dependency

[lessw2020]

Description

Updates softmax.py to be internally consistent with other tilegym kernels ala attention.py by removing the numpy import dependency and use -math.inf and ct.float32. (vs -np.inf and np.float32).
This also updates to use the ct dtypes (ct.float32 and ct.int32) to replace torch.int32 and torch.float32 to further standardize dtypes.

CI Configuration

config:
  build: true
  # valid options are "ops" and "benchmark"
  test: ["ops"]

Checklist

• [ X ] Code formatted and imports sorted via repo specifications (./format.sh)
• Documentation updated (if needed)
• CI configuration reviewed

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[hannahli-nv]

/ok to test fb27207

[xjmxyt]

/ok to test fb27207

[hannahli-nv]

LGTM, thx!