GitLab

## Major changes

1. Update the compatibility matrix for PyTorch upstream/ROCM Fork
2. Update `docs/How To Generate Tuning Database.md`
3. Refactor the `tritonsrc/` files so that there is only one kernel per
file
* `tritonsrc/bwd_split_kernel.py` -> `tritonsrc/bwd_kernel_dk_dv.py` and
`tritonsrc/bwd_kernel_dq.py`
* `tritonsrc/bwd_kernel_common.py` -> `tritonsrc/bwd_inner_dk_dv.py` and
`tritonsrc/bwd_inner_dq.py`

## Minor changes
1. Remove v1 directories, including `python/` and `tritonsrc/v1`
2. Remove fused kernel (`tritonsrc/bwd_fused_kernel.py`) since they are
not well tested
3. Remove unused files under `tritonsrc`
4. Remove unused CMake variable `AOTRITON_EXTRA_COMPILER_OPTIONS`