Check for uninhabited types in sub-patterns (#12609)

[canndrew]

This alters the exhaustiveness-checking algorithm for pattern matches to
avoid raising spurious errors about cases not being covered for
uninhabited types.

Specifically, the construct_witness function now returns an Option. If
it sees that DUMMY_WILD_PAT is being used to match on an uninhabited
type it returns None to indicate that there is no witness. We look for
DUMMY_WILD_PAT specifically and not any wildcard pattern so that
wildcard patterns explicitly written by the programmer continue to work
without raising errors about unreachability.

[rust-highfive]

r? @arielb1

(rust_highfive has picked a reviewer for you, use r? to override)

[bors]

☔ The latest upstream changes (presumably #36551) made this pull request unmergeable. Please resolve the merge conflicts.

[bors]

☔ The latest upstream changes (presumably #36857) made this pull request unmergeable. Please resolve the merge conflicts.

[arielb1]

I would prefer that people on the lang team to look on this.

[bors]

☔ The latest upstream changes (presumably #36695) made this pull request unmergeable. Please resolve the merge conflicts.

[canndrew]

Looks like the code that this touched went through a major refactor and this PR will have to be rewritten.

[steveklabnik]

/cc @rust-lang/lang @rust-lang/compiler

[nikomatsakis]

@canndrew shall we close the PR then and you can re-open when ready?

[canndrew]

@nikomatsakis Sure

[canndrew]

There's another, more complete and correct attempt at fixing this here: #38069