Fixes default kernel name.

Fixes pip dependencies.

xlstm 7b release

Make license Apache 2.0.

Lower the minimum required python version to 3.9

Fixes the correct setting of _num_blocks in the mLSTM block, which af…

…fects initialization. Fix (no) weight decay assignment for FSDP.

Fixes: Add sLSTM layer step(), typos, run instructions in Readme.md

Fix for non CUDA (cpu only) environments.

Initial release of the xLSTM code.