Some cleanups and add a solve_limited

[jackh726]

This could be two separate PRs, but the cleanups are small:

• Moved methods from InferenceTable onto Context directly, since they don't need an &self
• Made any_future_answer take a &self instead of a &mut self

The bulk of the PR involves adding a should_continue argument to peek_answer and next_answer, which allows stopping of solving before an answer is ready, when a QuantumExceeded has been returned. Right now, it's a argument-less function, so there's no real "feedback", so it's pretty much limited to either a time-based or call number-based implementation.
However, some ideas for how to extend this:

• Pass the current (or expected) substitution to the function. This could allow the caller to accept that an answer is "good enough".
• Be a bit smarter with the returned answer after false is returned. Right now, the returned answer is the current substitution as Guidance(Suggested(...)). Instead, we could take into account the possible answers (similar to how any_future_answer works). Or, maybe we just return None instead.

[jackh726]

So I'm not sure why CI is complaining. It builds/tests fine locally on stable 1.40.0. And the CI is complaining that Fn is a struct not a trait?

EDIT: because this wasn't rebased onto the current master, where Fn is a chalk-ir type