Squash llama2 into one commit to resolve conflicts

[JacksonKearl]

No description provided.

[cristianoc]

Looks great.

There are quite a few commented out lines that can be cleaned up if the current code is stable.

Also, some logic such as prompt truncation might belong to some general functionality rather than one specific kind of model, to avoid repeating similar logic in different places. Though that can be done later.

[cristianoc]

Nit: why is the .vsix file name shipped with a release not versioned?

[JacksonKearl]

@cristianoc the bulk of the llm/... edits are carried over exactly from the jesse/llama branch, not sure what @jesse-michael-han wants to do with those. We talked about how it'd make sense to flip the logic somehow so that the context-providing/prompting stuff is dependent on the model rather than the runtime.

In theory basically everything but the exact "executor" function could be handled more or less the same for all runtimes, and a map of some sort could be constructed to associate models with the prompting/context they need.