Update key benefits

[radka-j]

This is to be done once the PyTorch refactor of AutoEmulate is finished. At the very least we should:

• explicitly highlight the PyTorch backend (e.g., as part of or instead the easy integration benefit)
• we might also want to rethink the model groupings
  • at the very least, the models in the tool might be slightly different (we will likely drop support for some, maybe we will add support for others)
  • should we have a "deep learning" group and should this include GPs?
  • should GPs be their own section and we highlight which types of GPs we support)
• by this point we might have additional downstream tasks and we should ask whether these should be highlighted somewhere