Consider implementing Serialize/Deserialize for type Locale

[sffc]

We currently have an impl for LanguageIdentifier but not Locale. This discourages people from using Locale: #6810

In my opinion, we should implement serde on both or neither. Since we've had it on LanguageIdentifier since #89, we should probably just add it to Locale.

Please note that this is not just slapping #[derive(Serialize)]; the impl should look similar to the LanguageIdentifier impl, which does the same string serialize/parse that we currently recommend be done in userland.

CC @Manishearth @zbraniecki @xuxiaocheng0201

[Manishearth]

I'm in favor!

[robertbastian]

I vaguely remember that we purposely didn't want to implement this because it cannot be done in a zero-copy way. But I guess with enough documentation it's fine.

[Manishearth]

Yeah, overall I don't think we should care about that. Most clients do not need zerocopy deserialization of ICU4X types: they may need it for our data, but not necessarily our types.

We have a test that ensures we never stick non-zc types in our data; that ought to be enough to protect us from the footgun of having a conveniently serializable (but not zc) Locale type.

[sffc]

We won't use the impl in the ICU4X data pipeline, but it seems useful for clients. I think we should offer it, with a link to the zero-copy docs for clients who want something a bit faster than the default impl.