Infix operation desugaring should preserve original span

[rochala]

Found during debugging of #22566

Basically, we created new apply based on arg inside Parens that led to shorter span.

        case Parens(arg) =>
          Apply(sel, assignToNamedArg(arg) :: Nil)

[som-snytt]

Sample diff of my suggestion that moves the point

--- a/tests/explicit-nulls/neg/byname-nullables1.check
+++ b/tests/explicit-nulls/neg/byname-nullables1.check.out
@@ -1,4 +1,4 @@
--- Error: tests/explicit-nulls/neg/byname-nullables1.scala:10:6 --------------------------------------------------------
+-- Error: tests/explicit-nulls/neg/byname-nullables1.scala:10:12 -------------------------------------------------------
 10 |    f(x.fld != null) // error
    |      ^^^^^^^^^^^^^
    |      This argument was typed using flow assumptions about mutable variables

showing point (column of the error pos) is the op of the infix op, not start of left operand.

[som-snytt]

I'll look at the other pos diffs from my suggestion, to check if they also make sense.

I agree with the logic of the original change here, but can't explain why "composition of positions" is so hard. I understand my suggested change without thinking much.

[som-snytt]

I tried

apply.withSpan(Span(left.span.start, right.span.end, point = op.span.start))

at the end of binop. That corrects some points in errors, and corresponds to "preserve the span". I'm not sure where point comes from in this expression. (Trying to make position calculations "read" correctly in Scala 2 was also a challenge.) Here union preserves point of the receiver, which is correct; I'll post the diffs.

[som-snytt]

deceptively, Span(start, end) is not the same as Span(start, end, point = start) as I assumed for some reason. I have drilled down into synthetic spans for the "check unused" lint, but I really don't want to hold that arcana in my head. Positions should "just work", if we can figure out the best syntax.

(The doc for def Span does say which flavor is synthetic, but frankly, "pointless" does not say "synthetic" to me, unless I've been recently imbibing the Kool-Aid.)