Add an API endpoint to load the last-used model

[anon-contributor-0]

Adds an internal API endpoint to the OpenAI API that allows loading of the last used model.

This new endpoint would be particularly helpful for scenarios where VRAM management is necessary - a third-party application can ask text-generation-webui to vacate VRAM (e.g. with /v1/internal/model/unload), then quickly reload the model that was just active once some other task is done (for example, image generation). That technique is employed in the sd_api_pictures extension with AUTOMATIC1111's Web UI. This PR would allow other applications to perform the same technique with text-generation-webui.

/v1/internal/model/loadlast triggers the new models.load_last_model() function if it's POSTed to.

As a bonus, this also fixes a bug in models.reload_model() - it would fail since shared.model_name was set to None by models.unload_model(), meaning that reload_model() would then attempt to load None as a result. Setting it to the newly-added shared.last_model_name variable should fix that issue.

[anon-contributor-0]

@oobabooga, any chance you could have a look at this? It's a relatively quick PR, and it could allow for better flexibility for users on lower-end hardware.

[anon-contributor-0]

@oobabooga Checking in again - any chance you'd have a cycle to take a look here?