When trigger suggest for incomplete completions, use new kind

[ramya-rao-a]

This PR introduces a new trigger kind for the case where completion provider are asked for completions as the previous list returned was incomplete.

cc @mjbvz and @dbaeumer

[mjbvz]

@ramya-rao-a Code changes look good. Curious how this will be used though. Something with emmet?

[ramya-rao-a]

Yes, it helps in my current exploration of integrating emmet completions in the css completion provider. The css completions are "complete" ones and the emmet completions are "incomplete". Since we can only have 1 list coming out of the LS, this makes the list "incomplete".

Knowing that the current request is for completing a previous incomplete list, lets us re-use the last completion result and only update the emmet completions. This way, we don't need to re-parse the file contents.

Other extensions can be benefitted in the same way. We know for sure the C++ completions are also of the "incomplete" kind and can use this to optimize their completion process.

[dbaeumer]

Looks good to me. Ping me when merged so that I can adopt for LSP.