feat(ui): Implement the “fuzzy match room name” filter

[Hywan]

Address #1911.

This patch implements a new pre-defined filter for room list entries: room_list_service::filters::fuzzy_match_room_name.

The idea is that all filters can be build via a new_filter function, which produces a closure Fn(&RoomListEntry) -> bool + Send + Sync + 'static.

For fuzzy_match_room_name, I'm using fuzzy_matcher from the skim project. It's battle-tested and widely used via skim. However, it works on char, and no Unicode transformations are applied. So, looking for stf won't match Ștefan —even with smart case/ignore case— for instance, because S is different from Ș.

So before matching the pattern over the subject, we apply some transformations. I wanted to do some transliteration + normalization.
Ideally, I wanted to use icu4x. The transliteration API isn't implemented yet, so I'll just use the normalization API. However, those dependencies can be quite heavy. Looking inside our dependencies, we see that url is using idna, which uses unicode-normalization, it's a project from @unicode-rs, not from @unicode-org. unicode-normalization is slower than icu4x but it's already in our tree of deps, so let's use it. Let's measure and see if speed could be a problem here.

This patch then adds a filters::normalize_string to run some transformations (NFD + filter out combining marks).

Let's see if this approach suits for most common needs for all languages and cultures.

This patch doesn't implement the bindings for entries_filtered to FFI yet, it will be done in another PR.