Avoid more constraints in result if trying views

[som-snytt]

In canProfitFromMoreConstraints, balk if mode.is(ImplicitExploration).

The failing case was due to the constrainResult "up the stack" in adaptNoArgs, which drove type comparisons and completions, including a conversion. The FunProto context had a console reporter that emitted the errors encountered, not an exploring reporter.

The justification is that "more constraints" is motivated by arg inference, so if the arg is already involved in views, further information in the result is not needed, especially at the cost of spurious exploratory errors using a context that doesn't reflect the current mode.

Fixes #24192

[odersky]

Nicely spotted! I have just one small change request.

[odersky]

It seems to me that we should keep the second condition for needsUsing. We don't want to change behavior under -source 3.3.

[som-snytt]

Ah, I guess the || was a boolean typo for &&.

[odersky]

Ah, my bad. At this point I don't know anymore what was intended. Never mind, we can keep it as is.

[som-snytt]

I'll spend a few minutes to think of a counter-example. More tests would be nice, and the details of the PR are already receding into the fogs of time.

[som-snytt]

It now always takes needsUsing as true. Previously, the intention was just to avoid warning about requiring using for explicit context bounds; but the fix for "correctly show implicitNotFound message" relies on it as well. It's always OK to set the apply kind to using.

The latter fix is not needed when adaptNoArgs avoids the typecheck to elicit default args. A previous fix mostly does avoid it, except for the test case added to tests/neg/i19594.scala.

Maybe further tweakage is possible, but not pursued here.