Error handling, code style, and using notices for command responses

[lboklin]

Notices are specifically intended for uses such as automated messages, so I made it send notices whenever the reply is not generated by the LLM.

Because AIChat is not properly using exit codes to indicate that something went wrong, I added a check for empty outputs. I would submit a fix to AIChat directly, but after repeated attempts at contributing to that project I've decided against wasting my time on that.

The misc refactors are more opinionated, but there are several reasons to prefer match over early returns:

• it's easier to follow the flow and branching of execution
• it makes adding new cases easier and leads to cleaner diffs
• (not applicable in this case, but) it enables totality checks

I couldn't think of a good way to replace the break 'outer with a more functional approach due to the intricacies of borrow handling and the convoluted logic involved, so I gave up on that one.

[lboklin]

I think I understand why you would use error! over info!, though. There is some serious info spam from matrix_sdk that can only be silenced by setting the log level to error. What I do if I need to debug is to use println! instead. I wish I knew of a way to selectively silence such outputs.