Made ref pattern bindings correctly pick Deref or DerefMut

[Kimundi]

Closes #15609

[rust-highfive]

r? @brson

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

[Kimundi]

r? @nikomatsakis

cc @eddyb

[nikomatsakis]

r+ modulo the two nits

[Kimundi]

@nikomatsakis re the tests: I only realized after I wrote them that I did an unnecessary amount of duplication. If I'd used a Cell field instead of just a panic, it could probably all be one test that ensures it got never touched.

I'll change that, plus theLValuePref stuff.

[eddyb]

@nikomatsakis There is a subtle reason for not having struct LvaluePreference(pub ast::Mutability) to begin with, though it's not much of one: in the non-mutable case, immutable is actually not preferred, there is no preference.
May I suggest struct LvalueUsage(pub ast::Mutability)?

[Kimundi]

Fwiw, I just added the from_mutbl method, since there are only 3 places where the direct conversion is used, but ca 20 places where one of the two constants are hardcoded.

[Kimundi]

@nikomatsakis: Okay, I addressed both nits.

• There is now LvaluePreference::from_mutbl(), since that operation was less common that having one of the two as a constant.
• There are only two test files now, one for the issue scenario, and one testing all possible legal combinations of ref mutability and deref operator.

[nikomatsakis]

@bors r+ nice!

[bors]

📌 Commit d6b7ca0 has been approved by nikomatsakis

[nikomatsakis]

@eddyb true, the name is somewhat misleading.

[ghost]

Naming issue: the name of the function still implies it's a predicate function but now it returns an Option. Same below.

[bors]

⌛ Testing commit d6b7ca0 with merge 172cd83...

[bors]

☀️ Test successful - auto-linux-32-nopt-t, auto-linux-32-opt, auto-linux-64-nopt-t, auto-linux-64-opt, auto-linux-64-x-android-t, auto-mac-32-opt, auto-mac-64-nopt-t, auto-mac-64-opt, auto-win-gnu-32-nopt-t, auto-win-gnu-32-opt, auto-win-gnu-64-nopt-t, auto-win-gnu-64-opt