Refactor hasOnlyLocalInstantiation to use isLocalToCompilationUnit

[mbovel]

Follow-up on #19803.

[odersky]

LGTM

[mbovel]

In scala-parallel-collections, MiMa complains about AddingHashTable not preserving binary compatibility:

private[mutable] abstract class ParHashMapCombiner[K, V](private val tableLoadFactor: Int)
  ...
  private[ParHashMapCombiner] class AddingHashTable(numelems: Int, lf: Int, _seedvalue: Int) extends HashTable[K, V, DefaultEntry[K, V]]
  ...

Source

scala-parallel-collections: Failed binary compatibility check against org.scala-lang.modules:scala-parallel-collections_3:1.0.4! Found 1 potential problems (filtered 287)
[error]  * method this(scala.collection.parallel.mutable.ParHashMapCombiner,Int,Int,Int)Unit in class scala.collection.parallel.mutable.ParHashMapCombiner#AddingHashTable does not have a correspondent in current version
[error]    filter with: ProblemFilters.exclude[DirectMissingMethodProblem]("scala.collection.parallel.mutable.ParHashMapCombiner#AddingHashTable.this")
[error] Module scala-parallel-collections:1.0.4+28-7d0e41ae-SNAPSHOT is not binary compatible with version 1.0.4. You have to relax your compatibility intention by changing the value of versionPolicyIntention.
[error] java.lang.RuntimeException: Failed binary compatibility check against org.scala-lang.modules:scala-parallel-collections_3:1.0.4! Found 1 potential problems (filtered 287)

Logs

[hamzaremmal]

@mbovel I've disabled automerge for now to avoid having unexpected CI issues tomorrow morning.

[odersky]

This might be a blocker, unfortunately. We do get a different public interface.

[dwijnand]

I think it might be a bug in isLocalToCompilationUnit, it should be:

-      || accessBoundary(defn.RootClass).isContainedIn(symbol.topLevelClass)
+      || {
+        val boundary = accessBoundary(defn.RootClass)
+        val topLevelCls = symbol.topLevelClass
+        boundary != topLevelCls && boundary.isContainedIn(topLevelCls)
+      }

[dwijnand]

And if you want to be really ambitious, I think we should move the module logic into isLocalToCompilationUnit too.

So I looked at cls.is(Private, butNot = Module) to understand if the module part is redundant. I think the reason is that when a class mixes in the trait, it needs to instantiate the module even if it's private. So I assume there's some trait in the standard library (with Scala 2 emitted bytecode) that contains a private object that we need to correctly instantiate. And in Scala 2 it had an outer pointer, so we need to pass an outer-pointer to its constructor. So I think the logic is correct there, the only progression we could do is start dropping outer pointers when it's defined in Scala 3 instead of Scala 2 - but probably that's not worth it.