support Mamba(2) architecture

[CarloLucibello]

it would be nice to support the Mamba
https://arxiv.org/abs/2312.00752
and Mamba 2
https://arxiv.org/abs/2405.21060
architectures.

It is probably impossible to have efficient implementations without some heavy gpu kernel programming, but one could start with slow but correct implementations.

Some simple but not optimized code can be find here
https://github.com/johnma2006/mamba-minimal
https://github.com/tommyip/mamba2-minimal

[MartinuzziFrancesco]

Thanks for the implementation links, I was looking for something like that some time ago! I'll definitively add some modern rnn approaches soon