Prefer focus for missing arg position

[som-snytt]

The focus puts the caret at the bad application, instead of hanging in space where an explicit argument might have gone in an alternate reality.

The two benefits are pointing at an existing thing that is wrong and also shifting the messaging left.

Sample dramatic diff:

-- Error: tests/neg/i14025.scala:1:16 ----------------------------------------------------------------------------------
1 |val foo = summon[deriving.Mirror.Product { type MirroredType = [X] =>> [Y] =>> (X, Y) }] // error
  |                                                                                        ^
  |                ^

Further future enhancements could include the hypothetical args printed in Scala 2 messages, perhaps under -explain:

f(x, y, z)
^
The missing context arguments were expected as follows:
f(e1, e2)(x, y, z)
 ^^^^^^^^

Fixes #15630

[dwijnand]

I almost feel like start is better than focus? Ie rather than the change being:

 2 |  val x = implicitly[List[Boolean]] // error
-  |                                   ^
+  |                    ^
   |
 8 |  val value3 = M.pure("ola") // error
-  |                            ^
+  |                     ^

make it:

 2 |  val x = implicitly[List[Boolean]] // error
-  |                                   ^
+  |          ^
   |
 8 |  val value3 = M.pure("ola") // error
-  |                            ^
+  |                 ^

WDYT?

[som-snytt]

When you put it that way, I like my suggestion for future enhancement, just show me the signature with the holes.

I see there is already similar formatting:

   |I found:
   |
   |    FooT.OrderFFooA[F, A](FooT.OrderFFooA[F, A](/* missing */summon[Order[F[Foo[A]]]]))

[som-snytt]

I really feel like I'm not so stupid that it should take more than one or two rounds of tests to see which tests failed. I will endeavour [typelevel spelling] to figure out why it can't just summarize at the end and eliminate the scrolling.

[som-snytt]

Excerpt from my journal during this expedition:

Day 37, I moved the caret to the left, then all the tests expecting // error on a particular line number fail if the caret shifted to the previous line. Moreover, Pixie the test rig is coy about showing me her diff. I must be going mad.

[som-snytt]

@dwijnand I think my previous comment, that "an improved show-me-the-holes message" is ideal, but I maintain that this delta is an improvement. I'm still very bothered by carets pointing at whitespace.