Tree-sitter Performance

[WillLillis]

Here's a first pass on the tree-sitter performance changes I asked about in #39. I brought in the lsp_textdocument crate to help manage in-memory copies of the buffers being edited. Using this crate, text edits can now be communicated and applied incrementally, as opposed to copying over the entire document every time a change is made. I also co-located a tree-sitter tree in the global text store map so that each document now has a corresponding tree. With this in place, the trees can now be maintained incrementally alongside the in-memory buffers. Edits are applied eagerly to each tree (inside handle_didChange()), but the tree is only re-parsed lazily as needed (in get_position_from_lsp_completion()).

I also swapped out the protocol-related structs defined in lsp/src/handle.rs with their equivalents from the lsp_types crate, as the crate's structs have some extra information by more closely matching the lsp spec. Some of this extra info was needed to interface with the lsp_textdocument crate, but using the crate's types should also make things easier to maintain overall.

I made some assumptions about how this should be implemented, so if there needs to be any changes/ refactoring I'm happy to work through that :)

Apologies that this PR is a bit on the long side, but the tree-sitter changes had to touch a couple different parts of the code base.

Closes #39

[ThePrimeagen]

i am scared to merge this :)

i'll have to pull it into my editor and really consider this

[ThePrimeagen]

also, merge confwicts pweeze

[WillLillis]

Many complexity spirit demon maybe hiding in this PR, but Tom (a genius, btw) said LGTM :)

In all seriousness though, if this isn't it I'm happy to refactor/ start fresh. If it would help to break this into smaller PRs that's fine too. Really appreciate you taking a look!