Flip default of diagnostic experimentality

[Veykril]

I wonder if it would help to invert this, so non-experimental diagnostics have to be explicitly marked as stable? IMO all diagnostics should start out experimental, and "stabilizing" them should at least require that they don't have false positives on some number of code bases.

Originally posted by @flodiebold in #19670 (comment)

[ChayimFriedman2]

I support this. I also think that we can (should) have a CI pass that runs diagnostics on some codebases when a diagnostic is stabilized, and see if there are any diagnostics of the stabilized error code.

[Veykril]

Oh ye that sounds like a nice idea