Content with inline comments mixed with formatting sometimes fails to parse

This is caused by "end format" events being sent to the inline annotation reader without matching begin format, which can happen because we are catching events at the middle of a line and while in XML everything should be properly nested, wikimodel can decide to call endFormat() when it sees fit.

Since we are only concerned with getting the plain text while parsing an inline annotation, begin/end format events can simply be ignored.

This is a regression introduced with CONFLUENCE-512.