job_chat: process long logs

[josephjclark]

At the moment our strategy for handling logs is super naive

Lightning sends logs up to apollo, and apollo dumps the logs into a prompt

But the logs could be huge!

Data is a particularly big problem - large JSON objects can take a lot of space.

We'll need to work out a summarisation or truncation strategy

• Remove duplicate lines (in apollo or lightning?)
• summarise the logs with AI (we should forward the question for some context - because summarise for what?)
• set a character limit on how many logs we send (crude but it'll prevent problems)
• Maybe we also need a UI strategy like only sending the logs the user has selected/flagged/marked

[josephjclark]

Probably two issues here:

1. A short term stability fix
2. A bigger feature to process logs really smartly (call out to another model like gemini to process long logs)