Add path completion support inside require calls

[paracycle]

Motivation

I got inspired by the recently released Markdown Language Server doing completion for references to other files. I realized that we can match require strings against a list of files in the $LOAD_PATH and suggest completions like that.

Note: This is still missing some of the functionality, but since it is gated behind a feature flag that should be turned on by the VSCode extension, I think it is safe to land. The things that are missing are:

1. There is no support for requires that start with a . or .. which make them relative requires.
2. There is no support for load paths that may be added by the application. For example, Rails applications add Rails.root / "lib" and Rails.root / "app/models", etc to the load paths, which this extension will not be able to see (since we don't boot the application we are running inside).
3. There is no support for require_relative.

Implementation

1. Register a completion provider that is triggered on the / character
2. Handle the completion request and process it if (and only if) it is originating from the string argument of a require call.
3. Match the completion prefix to the list of paths in $LOAD_PATH and offer completions.

Automated Tests

Added a small automated test

Manual Tests

1. Open the https://github.com/Shopify/vscode-ruby-lsp project.
2. Add a

           "completion": {
             "description": "Enable code completion features",
             "type": "boolean",
             "default": true
           }
   feature inside the package.json file and a:

            "completion": true
   default
3. Run the project
4. Open the ruby-lsp project in the Extension Host window.
5. Start typing some requires and see autocompletion for paths.

[vinistock]

This is awesome ❤️ . I wonder if it would be worth finding files in the current directory as well? For example, if you create a brand new file and want to require it, it won't be in the load path yet.

[paracycle]

@Morriar and I reworked this PR so that it uses a more efficient Trie (which is also called a Prefix Tree) data structure to index files in the load path and brought it up to work against the changes in latest main.

@vinistock Currently we reindex the files on the load path for every request to completion, so we would be able to discover any newly added files. I know this is not the most efficient way of doing it, but based on my usage testing the responsiveness seems to be just fine for now.

[vinistock]

This works really well! Some comments about the code structure

[vinistock]

Amazing!! 🚀