Add a check against multiple `only-` directives in the same line

[scottmcm]

In 556b47e I wrote

//@ [OPT3LINX64] only-linux only-x86_64

which based on #123185 (comment) appears like it didn't run on just x64.

Assuming I'm right that this is what happened (if not, please close this) then it would be nice for the same helpful checker that tells me I did it wrong when I write linux-only to also tell me that I can't put multiple only checks on the same line.

[jieyouxu]

The reason is that I believe compiletest directive parsing for only-* assumes that the directive takes the form

//@ only-<condition> [comment...]

where anything following only-<condition> is treated as an optional comment

[jieyouxu]

We are able to detect unknown directives on their own, because we just have a non-exhaustive list of known directives and match against that. I suppose we could just try a match against the same known directive list for only-* comments to point out that you cannot chain directives in the same line? Although if we're doing it for only-* comments, what about other directives? It's a bit tricky because IIRC some directives allow comments, other directives don't accept comments...?

[scottmcm]

Brainstorming: maybe it'd be reasonable to require that the comment be in parens

//@ only-<condition> ([comment...])

or

//@ only-<condition> because [comment...]

or something like that?

[jieyouxu]

It is reasonable, the problem is that we have a lot of pre-existing directives that don't take this form (and we'd have to do another massive migration pass). I'm also not sure what ui_test does about this.

[jieyouxu]

This would've been a less surprising directive grammar from the beginning, but of course hindsight is 20/20. I'm not sure it's worth the churn to change the grammar now.

What we can do now without churn, however, is to explicitly document that directives are intended to be one-per-line in the dev guide (rust-lang/rustc-dev-guide#1962).

[Urgau]

Without changing the syntax, we could probably still error out if the sentence right after the directive is a well known directive. That way we should be able to catch @scottmcm mistake without massive change.