since tokens already have associated context information, we should be able to provide context-aware completions in the language server.