Implement Mixed-Curvature Transformers

[pchlenski]

Stub classes for mixed-curvature attention:
This would entail implementing all of the following:

• StereographicLayerNorm
• StereographicAttention
• StereographicTransformer

Reference code:
There is a pre-existing implementation of sectional curvature in the supplemental materials of the pCurve Your Attention OpenReview submission](https://openreview.net/forum?id=AN5uo4ByWH). The relevant implementations are in code/encoders/FPST/networks.py. The class mappings are:

• StereographicLayerNorm --> StereographicLayerNorm
• StereographicAttention --> StereographicAttentionLayer
• StereographicAct --> implemented in Manifold.apply()
• StereographicLinear --> Implemented in KappaGCNLayer with A=None
• StereographicTransfomer --> StereographicTransformer

Paper:
See Cho et al (2024), Learning Mixed-Curvature Representations in Product Spaces for a description of the method.

[pchlenski]

Implemented in #18 H/T @rrkhan