⚡️add florence+linear action head policy, and adapt it to pusht env

[StarCycle]

What this does

add florence+linear action head policy, and adapt it to pusht env

How to checkout & try? (for the reviewer)

Examples:

pip install -e ".[florence]"
python lerobot/scripts/train.py policy=florence env=pusht dataset_repo_id=lerobot/pusht wandb.enable=True

TODO

1. Tune the performance for pusht
2. Whether it satisfy the coding format of LeRobot?
3. adapt to aloha-gym env
4. add diffusion action head (Chi Cheng's diffusion transformer & DiT developed by @mbreuss at KIT)