Improve indent handling in parens

[som-snytt]

For outdent in parens, comma must be at end of line if the indented region was introduced by a colon.

Fixes #22527

[som-snytt]

That breaks sudoku!

      fold((accu: Int, n: Int) =>
           if (invalid(0, x, y, (n + 48).toChar)) accu else {
             m(y)(x) = (n + 48).toChar;
             val newaccu = search(x + 1, y, f, accu);
             m(y)(x) = '0';
             newaccu}, accu, 1, 10)}

which is hard to read and misleading indentation, but is it illegal.

The intuition here would be that if you indent the body of your function literal, you intend it to look like a block, and not mashed together with trailing args.

Restricting the restriction to COLONeol might be intuitive enough, or may be just ad hoc.

-              // For some region prefixes (COLONeol, EQUALS) only OUTDENT if COMMA at EOL
-              if canStartIndentTokens.contains(r.prefix) && !statCtdTokens.contains(r.prefix) then
+              // For region prefix COLONeol, only OUTDENT if COMMA at EOL
+              if r.prefix == COLONeol then

Maybe that is OK, because the colon is special.

I noticed that we wanted templates defined with =, and we wanted = to be :=, and wound up at :.

[som-snytt]

That breaks scalaz!

  def applyLens[A, B](k: B => A)(implicit e: Equal[A]): Store[A, B] @> B =
    lens(q => {
      val x = Need(q.pos)
      val y = Need(q put x.value)
      Store(b =>
        Store(w => if(e.equal(x.value, w)) b else y.value, x.value), y.value)
    })

where it looks like Store(b => body) because of the indentation.

[som-snytt]

The spurious timeout.

[som-snytt]

unindented args complains about

            addParamssText(
              addParamssText(keywordStr("extension "), leadingParamss)
              ~~ (defKeyword ~~ valDefText(nameIdText(tree))).close,
             trailingParamss)

That blocks other corpus tests. Probably it is too opinionated, but I'll try unblocking just to collect the data.

[som-snytt]

Rebased only. Not sure if anyone has an opinionated opinion about this syntax niche, whether it's better to support it or defund it.

[som-snytt]

I verified that tests passed in February commit
d194a49

so now I have the pleasant task of finding out what changed since winter.

Edit: yes it was me. The fix to infer indent after old-style conditional means it can't be combined with the current tweak, which is just as well.

#22611

Worth adding that it took me a while to realize that the test was using old-style syntax; in fact, old and new side-by-side.

[som-snytt]

This unimproved error is Illegal combination of named and unnamed tuple elements.

def g: Unit =
  identity(
    x =
      class X extends AnyRef, Serializable // error
      27 // error
  )

that is

      identity[(Unit, Any)](
        Tuple2.apply[Unit, Any](
          {
            class X() extends AnyRef() {}
            ()
          },
        Serializable)
      )