feat: simple transformer embedding net (#1494)

* attention and rotary encoding

* rm unnecessary parameters

* standard rope

* implemented transformer architecture

* fixed the rope frequencies

* added support for MoE (from Mixtral)

* comments and vit embedding

* added arxiv ref

* final fixes

* type linting

* fixed types

* type fixes

* type fixes

Author: NicolasRR

sbi/neural_nets/embedding_nets/__init__.py

@@ -10,6 +10,7 @@
     ResNetEmbedding1D,
     ResNetEmbedding2D,
 )
+from sbi.neural_nets.embedding_nets.transformer import TransformerEmbedding
 
 __all__ = [
     "CausalCNNEmbedding",