[ConstraintSystem] Attempt to select disjunctions that split constrai…

[rudkx]

…nt systems.

When selecting the next disjunction to attempt, try to find one that
is a disjunction of bindings where the type being bound is the
converted-to type in a conversion constraint. Attempting these early
makes it possible to split constraint systems, eliminating exponential
behavior.

Fixes: rdar://problem/40344044

[rudkx]

@swift-ci Please smoke test

[rudkx]

@swift-ci Please test source compatibility

[xedin]

This seems reasonable, also makes me wonder if we should prioritize disjunctions related to implicitly unwrapped optionals in general because they give more context just like coercions?

[rudkx]

@xedin I am not sure if this is what you mean, but one thought I had when I was doing the IUO work is that we should probably prioritize any disjunction that consists strictly of bind constraints in the same way we try to find a type variable we can bind before looking at disjunctions, since attempting types tends to cause us to be able to fail faster.

I'll try pushing a PR with a change to prefer those if none linked to a conversion is found and see if it helps (or hurts) any of the existing tests we have in the validation suite.