Support string 'ID' Fields

[ggilmore]

I noticed a comment here: https://github.com/rust-lang-nursery/rls/blob/a793188fd8021ff6ec2779d6f8fa3dea3bc13ad4/src/server/message.rs#L116-L138, but I thought that I'd make an issue to track it.

LSP supports arbitrary strings as the ID for a given request, but numbers are the only supported type in RLS. It'd be great if RLS were able to fully adhere to the protocol in this aspect.

[nrc]

This was previously discussed in #360. Note that we do support, e.g., "42" as an id (which we believe is the intent of the spec) as well as 42, but not "foo".

I'd not be opposed to supporting all strings, but as far as I know this is not supported by any clients, so it is so low priority that it is not worth tracking. If it ever becomes a real issue, I'd be happy to track it then.

[keegancsmith]

The spec does not allude to the ID being a number, only that the response JSON uses an identical id field http://www.jsonrpc.org/specification#request_object Discussion on the JSON-RPC

FYI Sourcegraph is a client of LSP and we use arbitrary strings as ID's.