Standalone remote client

[gmertes]

Remote execution no longer needs a local copy of trained model assets. Checkpoint data (and any missing local model parameter) is fetched from the server. Tested and working for all model plugins.

Also supports remote patch_retrieve_requests, the server will send back the patched request if the model requires it.

Changes:

• Adds a new model class RemoteModel that handles the remote model run and parameter fetching
• Add a tqdm progress bar during remote inference
• All the remote code now lives in its own subpackage remote
• Create an empty remote config file during pip install at ~/.config/ai-models/api.yaml
• Remote models can be queried with --models --remote