I digged a little deeper. The test case is a great one
The DocLexer cuts ut -1 position from the first @ (which is the @ in the @example.com). so that the lexer has to process:

o@example.com"
     * }
     *
     * @AnnotationTargetPropertyMethod("Bar")

So that the catchable pattern: ("(?:[^"]|"")*") matches here to greedy (it matches string in quotes):

"
     * }
     *
     * @AnnotationTargetPropertyMethod("

As a result the lexer does not catch the correct @ from the Annotation

I could not think of a fast fix for this. But maybe tomorrow.
(leaving quoted string pattern with an @ was a wrong idea)

Its not a workaround to do more or less cutting in the parser, because not-well-formed quoted strings would break anyway

can someone think of a case, where a string (in some annotation) has a newline in it? Otherwise the tests do not fail, when I leave the string matching pattern with a newline.
I still had another idea to restrict the position of a annotation more strictly to:

• the beginning of a new line (\s**\s* before)
• in another annotation

but the second is not solvable with the lexer itself and would transfer the quoted string matching to the parser, because it needs context.

I'll let seldaek pull, what I have (is this the right way?)