Optimize the use of Boyer-Moore searching

Searching and comparison operators should always be noexcept, which is a problem with our boyer-moore algorithms, because they may allocate. std::boyer_moore_searcher is also not noexcept.

So we need to write new private BM matchers, which should separate the table generation from the table use, so as to decouple the charset used for building BM tables from the one used to match against. std::boyer_moore_searcher is nice API that way, since it it completely customisable by passing user-defined hashers.

As regards QTBUG-100239, we need to cap the stored case-folded needle, so it can be stored in a statically-sized buffer. Maybe something like 256 chars or bytes. The result of matching using BM are then only candidates which, if the needle is longer than 256, need to be confirmed by a sliced().startsWith(needle.sliced(256)).