Another given priority scheme to ensure transitivity

[odersky]

It applies the following tweaks:

• Do treat givens as higher priority than implicits. The previous logic was faulty, but is corrected now.
• Drop special handling of NotGiven in prioritization. The previous logic pretended to do so, but was ineffective.
• Do use the old priority scheme for implicit/implicit pairs.
• Do use owner score for givens as a tie breaker if after all other tests we still have an ambiguity.

[odersky]

Alternative to #21305

[odersky]

@WojciechMazur Can we test this one as well with the OpenCB?

[WojciechMazur]

I've started the OpenCB for 3.5.0-RC6 + this change so we can compare it directly with the results from alternative PR and don't need to care of other issues present on main. I should have results in ~3h

[WojciechMazur]

There seems to be less arnings in general when compared to previous solution

Given stats for 3.5.0-RC6 (baseline)
Given alternatives choice: Some(the second alternative) - 483
Given alternatives choice: Some(none - it's ambiguous) - 762
Given alternatives choice: Some(the first alternative) - 264
Projects with changed alternatives warnings - 23

Given stats for 3.5.0-RC6 + #21305
Given alternatives choice: Some(the second alternative) - 1868
Given alternatives choice: Some(none - it's ambiguous) - 21
Given alternatives choice: Some(the first alternative) - 831
Projects with changed alternatives warnings - 88

Given stats for 3.5.0-RC6 + this PR 
Given alternatives choice: Some(the second alternative) - 1437
Given alternatives choice: Some(none - it's ambiguous) - 22
Given alternatives choice: Some(the first alternative) - 215
Projects with changed alternatives warnings - 28

There is 1 new ambiguous choice in com-lihaoyi/upickle (it was not ambiguous in neither of other revisions)

Search [1]: ReadersVersionSpecific_this.Reader[upickle.Gadt2.IsDir[V]] : choice=none
    -  (upickle.Gadt2.IsDir.rw : [V²] (implicit evidence$1: upickle.default.ReadWriter[V²]): upickle.default.ReadWriter[upickle.Gadt2.IsDir[V²]] )
    -  (upickle.default.superTypeReader : [T, V² >: T] (implicit evidence$1: scala.deriving.Mirror.ProductOf[T], evidence$2: upickle.default.Reader[V²], evidence$3: scala.deriving.Mirror.SumOf[V²], x$1: scala.util.NotGiven[upickle.core.CurrentlyDeriving[V²]]): upickle.default.Reader[T] )

Here're links to each of the types:
ReadWriter
default.superTypeReader
Gadt2.IsDir and it's rw implicit instance

No new compilation failures in OpenCB, lemonlabsuk/scala-uri is now building successfully (failed previously)

[odersky]

There is 1 new ambiguous choice in com-lihaoyi/upickle (it was not ambiguous in neither of other revisions)

OK, here's what goes on here.

• superTypeReader is a given and isDir.rw is an implicit.
• Both superTypeReaderandisDir.rw` are polymorphic
• superTypeReader's type variables can be instantiated to match a supertype of isDir.rw's type. This would normally
  make isDir.rw chosen over superTypeReader except that isDir.rw is now an implicit and therefore loses the comparison. The result is that isDir.rw is not better than superTypeReader in the "wins type" contest.
• In the other direction we alsp get a "not better" since isDir.rw's type variable cannot be instantiated to match superTypeReade's result at all.

[WojciechMazur]

Based on the results in #21328 should I run additional OpenCB build with additional 5d58407 change?

[odersky]

No I think we should hold off with l 5d58407 unless the other PR #21328 is disappointing. I think the other PR is both nicer and might solve more problems.

[odersky]

Superseded by #21328