I think this is just a picky optimization.
The first one runs the constructor to instantiate a new string, then gets its class (which is presumably a static property anyway). The second doesn’t have to run any constructor and just grabs the static class name from the type.
Maybe there’s more implementation nuance here but it seems like an opinionated rule that has zero effect on performance unless that code is being called thousands of times every second. And even then the compiler probably optimizes them to the same code anyway.
I had the same problem and eventually abandoned it. Even though I had good results with it finding content, the book metadata is usually terrible and I end up having to manually fix it in Calibre and then force download metadata and cover art.
If I’m already having to do that anyway, might as well just acquire the book manually and import it myself.