Fix type inference involving wildcards

[adriaanm]

Use == instead of =:= because we're looking for occurs, not equality.

When we were using =:= to detect occurrences of type params in bounds,
we'd accidentally unify too much (e.g. on wildcards / other type vars).

Discovered while looking into why this doesn't type check:
java.util.stream.Stream.of(1,2,3).map(_.toString)

I noticed that there was cross talk between the type params of java.util.Function during type inference.

This partially regressed in #6492, but was actually wrong since the one with *** empty log message ***

Fix scala/bug#11558
Relates to scala/scala-dev#184

[retronym]

Confirming that the problematic example from scala/scala-dev#184 progresses here.

Before:

scala> "abc".chars.boxed.map(x => x.toString).collect(java.util.stream.Collectors.toList())
                               ^
       error: type mismatch;
        found   : Integer => String
        required: java.util.function.Function[_ >: Integer, _]

After:

scala> "abc".chars.boxed.map(x => x.toString).collect(java.util.stream.Collectors.toList())
res4: java.util.List[String] = [97, 98, 99]

[retronym]

Standalone test case that progresses:

trait F1[A, B] { def apply(a: A): B }; trait S[T] { def map[R](f: F1[_ >: T, _ <: R]): S[R] }; (f: S[Int]) => f.map(_.toString)

[retronym]

Pulling a few more threads in https://github.com/retronym/scala/tree/review/8129

[adriaanm]

@retronym I incorporated a few strands of your review branch. LGTY?

[SethTisue]

regression in overload resolution: scala/bug#11755