Support that client and server agree on the word boundaries of a language

Filtering of completion items in the simple insert text case is based on word boundaries of the language (see https://github.com/microsoft/language-server-protocol/issues/898#issuecomment-593968008).

Since completion items are generated by the server they both must have the same understanding of what a word boundary is. I see different possibilities to address this:

1. standardize on the word boundaries. IMO not a good solution since it will be hard to achieve
2. client retrieve word boundaries from the server. The problem with this approach is that basic editing experiences require to start up a server. Especially in remote cases this can cause quite some problems
3. client tells server about its word boundary configuration. This can be achieved in two ways:
   1. in the intialize call
   2. using a getConfiguration call send from the client to the server.

I am actually in favor of the 3.ii.

Start by reading the linked language-server-protocol issue 898 comment and the three approaches described here, especially the proposed getConfiguration path. Resolve which protocol mechanism should carry word-boundary configuration, document the decision, and confirm that both client and server can use the same boundaries for completion filtering.