compiletest: technically potential false positives for filecheck annotations

[jieyouxu]

This is a fun one: in #121561 I added checks to error if we detect any legacy-directives-like comments (which are of the form // <known_directive_name>.

In codegen tests, we have filecheck directives that can be revisioned:

//@ revisions: known_directive_name
// known_directive_name: noundef

... but unfortunately, // known_directive_name looks like a legacy directive to compiletest so we might error here even though it is not actually a legacy directive (it just "looks" like one for the heuristic I added).

Technically a bug (false positive), but in practice changing the revision name to something else would sidestep this check.

Note: the legacy directive check would also fire on

// this is a block of comment that has a
// ignore in the middle of the sentence
// that looks like a legacy directive

[jieyouxu]

Realized this in #125813 (comment), the long term solution is probably to have compiletest directive handling be cooperative between "base" directives and test suite "customized" directives.

[workingjubilee]

fwiw despite noting the issue, I think the prefix doesn't matter too much: I reviewed FileCheck's code (as much as I understood it, anyways) and it seems FileCheck is very simple. I don't think it is aware of any comment styles itself. I think it literally just scans for the prefix on a line and assumes you won't introduce code that resembles a FileCheck directive.

[jieyouxu]

compiletest triage: I deleted the legacy directive check so this is no longer a problem.