Rustdoc shouldn't collect all attributes into a single string when parsing markdown

[jyn514]

Unfortunately, the API seems pretty limited: https://docs.rs/pulldown-cmark/0.7.2/pulldown_cmark/struct.Parser.html

So until it's being extended, I guess we're kinda stuck with this extra allocation...

Originally posted by @GuillaumeGomez in #75665

[GuillaumeGomez]

To be checked but I *think* we could extend their API to allow them to take an iterator over char or something along the line.