feature: handle unavailable errors

[kerma]

Did you check the docs?

• I have read all the docs

Is your feature request related to a problem? Please describe.

There are probably a various errors which fall into the "AI API not available" bucket. These could be handled gracefully, by having an automatic retry in place, a way for a user to easily retry, or fall back to another provider.

# assistant:
Stream Error: {"type":"error","error":{"details":null,"type":"overloaded_error","message":"Overloaded"}   }
Error: {"type":"error","error":{"details":null,"type":"overloaded_error","message":"Overloaded"}   }
    at new AnthropicError (unknown)
    at new APIError (/Users/user/.config/nvim/plugged/magenta.nvim/node_modules/@anthropic-ai/sdk/error.mjs:7:9)
    at new APIConnectionError (/Users/user/.config/nvim/plugged/magenta.nvim/node_modules/@anthropic-ai/sdk/error.mjs:70:9)
    at generate (/Users/user/.config/nvim/plugged/magenta.nvim/node_modules/@anthropic-ai/sdk/error.mjs:33:24)
    at <anonymous> (/Users/user/.config/nvim/plugged/magenta.nvim/node_modules/@anthropic-ai/sdk/streaming.mjs:50:40)
    at processTicksAndRejections (native:7:39)

Describe the solution you'd like

A reasonable delay (5s) between retries would probably be the best UX, together with some indication to the user what's happening.

Describe alternatives you've considered

Capture the errors and show simpler "try again later" message to the user.

Additional context

No response