Add a custom context/token limit for models.

[root-reindeer-flotilla]

With the release of Gemini 2.5 preview, the cost isn't as bad as some other models, but it really eats tokens if you use the Boomerang method, where each subtask has to send a lot of tokens. Or just dealing with a very large context.

I suggest, if it's possible, to add a slider that we can limit what Roo thinks the token context window size is. Gemini 2.5 has a 1 million context window, but like I said, that really adds up very fast. Obviously this would reduce the effectiveness, but it would be a good option to help control costs for smaller projects or tasks. And of course this would only go down, you can't increase the context limit on our end, don't think that's what I'm asking for.