Bump to v1.5.0

[CI] Hard-code pytorch-triton version for torch 2.6.0.dev20241001

Actually change torch 2.6.0dev version

[CI] Change torch 2.6.0.dev20241010 to dev20241001 for nvcr 24.11

[CI] Set wheel name with cu12 instead of cu124

[CI] Don't include ATen/cuda/CUDAContext.h to avoid cusparse.h

[CI] Install libcublas and libcusparse

[CI] Only install nvcc

[CI] Update typing-extensions for Python 3.13 build

Bump to v1.4.0