Feature request: load/cache several models in RAM

[green-anger]

Extracted from this conversation.

When you change a model (checkpoint) it must be loaded from the drive every time. If there was capability to pre-load multiple models, it'd be possible to switch between them much faster reducing delay between generations.

UX suggestions:

• in the model selector list add an indicator next to each item to quickly see if it's in RAM or must be loaded
• add a button/link to call a list of all models where you can load/unload one by one or in bulk

NB: I haven't used Swarm yet, so the above is not related to the current or planned UI.