Add stellar tutorial and refactor AION forward generation for improved ease of use

[lhparker1]

This pull request introduces updates to the README.md file and significant changes to the aion codebase, focusing on modality definitions and the model's forward pass logic. The most notable changes include adding num_tokens attributes to various modality classes, refactoring the forward method in the model to dynamically compute encoder and decoder tokens, and improving the documentation.

Documentation Updates:

• README.md: Added a centered AION logo image for better visual presentation and renamed the "Key Use Cases" section to "Example Use Cases" for clarity. [1] [2]

Modality Enhancements:

• aion/modalities.py: Introduced a num_tokens attribute to all modality classes (e.g., HSCImage, LegacySurveyImage, DESISpectrum, GaiaFluxG, etc.) to define the number of tokens each modality contributes. This enables dynamic token handling in the model. [1] [2] [3] [4] [5]

Model Forward Pass Refactor:

• aion/model.py: Refactored the forward method to dynamically compute num_encoder_tokens and num_decoder_tokens based on input and target modalities. Added a helper _forward method for modularity, and replaced target_mask with target_modality as an input parameter for better usability.

These changes collectively enhance the flexibility and maintainability of the codebase, particularly in handling diverse astronomical modalities and adapting to varying input/output configurations.

[EiffL]

OK LGTM