chat: fix issues with quickly switching between multiple chats

[cebtenzzre]

If you switch between chats too quickly, it can result in the UI getting into inconsistent states (e.g. a progress bar stuck at 60%) and sometimes seemingly deadlocking (which happens during the new "Waiting for model" status).

To fix UI inconsistencies:

• Move trySwitchContextInProgress, modelLoadingPercentage, and isCurrentlyLoading properties from the window to the chat so the load status always reflects the current chat
• Initialize load status at the beginning of loadModel instead of only in setModelInfo - reloadModel calls this too
• If loadModel is cancelled, quickly return instead of trying to fall back and reporting an "invalid model" error
• On load failure, instead of showing only the bottom "reload" button, show neither

To prevent quickly switching chats from queueing many slow operations:

• Skip context switch if an unload is pending
• Skip unnecessary calls to LLModel::saveState

UI enhancement:

• Show "waiting for model" as a separate state when switching context, so it's clearer what is taking so long

Other bugfix:

• Use std::unique_ptr in the LLModelStore and fix a related memory leak - not absolutely necessary for this PR, but it makes some other changes slightly simpler

Non-functional changes:

• Use std::optional to store zero or one items in the LLModelStore
• Remove m_shouldTrySwitchContext, which didn't do anything useful, and simplify the related code

[manyoso]

This is tricky code. Let's do a huddle for review on this one.