experiment with resurrecting the recursive solver

[nikomatsakis]

We should experiment with returning to a "recursive style" solver in chalk. As described in this gist, recursive solving does offer some potential advantages:

• if we extend the setup, it may resolve the ambiguous associated type problem (ironic since we previously removed the recursive solver in part because we couldn't make it handle associated types well)
• less overall work, at the cost of completeness
• cleaner handling for multiple threads doing queries at once without locking (the cache is separated from the main solver stack)
• smoother integration into systems like chalk

Work items:

• Initial PR
• Coinduction semantics coinduction semantics in the recursive solver #399
• Invoke program_clauses_for_goal on the canonicalized form vs instantiating the goal first (comment)

This issue has been assigned to @flodiebold via this comment.

[nikomatsakis]

@flodiebold is working on this

[jackh726]

@rustbot assign @flodiebold

[jackh726]

Dang it that didn't work.

[nikomatsakis]

@rustbot assign @flodiebold

[jackh726]

I would say that the "experiment" is done. Any follow up issues (including #399) can just be separate.