Adjust print_response to separate error-printing from result display, remove loop

[BioTurboNick]

Took a stab at fixing #59209 , but it could be introducing other ways it could fail that I'm not appreciating?

1.11.4 logic:

1. If it's an error, print the error then break and return
2. If it's an error, and printing the error causes an error, catch it, try two levels of fallback printing, then break and return
3. If it's okay, display the result then break and return
4. If it's okay, and displaying the result causes an error, print a message, rethrow, catch, populate the response variables with the new exception info, and start the loop over to follow the error printing pathway for those values

In no case does the loop execute more than twice.

In 1.11.5+, condition 4 ends up following path 2, and path 4 would only be followed if there's an exception in the scaffolding around the display invocations.

I restructured it to take out the loop, now it just attempts display if there was no error, accounts for errors during display, and then errors from either response or display are printed.

I'm not sure if I used the sigatomic_end calls correctly. EDIT: See comment below

Also tweaked the final fallback message because it suggests the issue is that the exception per se is a nested type, rather than that printing one exception is triggering another exception. And added a condition to trim REPL frames when it originates from within print_response.

Fixes #59209

[BioTurboNick]

The 1 CI failure seemed to be an unrelated timeout or similar, trying to merge/rerun one time before marking for review....

EDIT: All green

[BioTurboNick]

I think I understand sigatomic_ now. The bracketing _begin and _end place everything in the function into an uninterruptable state so that the primary functions of the REPL happen in an orderly fashion. However, we want to temporarily allow interruption during potentially long-running code, like display of a result or processing/printing a stacktrace. So the atomic region is ended just before those sections, and either restored after by a _begin or invisibly restored by the try block ending. Everything in the catch block operates in the sigatomic context of the enclosing block.

[BioTurboNick]

This would be good for 1.12 backport. Could backport to 1.11 too, though may be a judgment call.

[vtjnash]

I think the Base.sigatomic_end stuff is some legacy baggage that keeps getting copied here by accident, but the change seems to look right (and work right on my machine in brief testing)