Subst canonical environment clauses

[jackh726]

Fixes #670

I'm not sure if this is the right approach. Recursive solver gets substed later because clauses get instantiated later. There's also a clone there which would be nice to get rid of.

[jackh726]

Actually, there's more wrong here that the test case doesn't cover.

[jackh726]

Better repro:

#[test]
fn env_bound_vars() {
    test! {
        program {}
        goal {
            exists<'a> {
                if (FromEnv(&'a ())) {
                    WellFormed(&'a ())
                }
            }
        } yields[SolverChoice::slg_default()] {
            "Unique"
        }
    }
}

[jackh726]

Okay this is good to go now

[jackh726]

Ugh this still isn't completely correct. This fails when the environment with bound vars from the Canonical end up in constraints.

I don't have a min repro yet, but it occurs in the chalkify tests in rustc

[jackh726]

Alright so, this is kind of a difficult design problem: what are the environments in Constraints supposed to mean. I've opted for now to just always pass an empty env to constraints, since it's unused currently in both rust-analyzer and rustc (and is blocking rustc integration upgrade). But this is a short-term solution.

I also don't have a repro currently.

[bors]

☔ The latest upstream changes (presumably #673) made this pull request unmergeable. Please resolve the merge conflicts.

[nikomatsakis]

@bors r+

[bors]

📌 Commit 09c9adc has been approved by nikomatsakis

[bors]

⌛ Testing commit 09c9adc with merge 1329e3a...

[bors]

☀️ Test successful - checks-actions
Approved by: nikomatsakis
Pushing 1329e3a to master...