Command ranking using both commands and titles

[forrestbao]

Previously we only match commands with queries. Now we can extract command titles, so we should leverage both.

1. Storing the embeddings of both commands and title. We will use three objects:
   • List[str], the commands themselves
   • 2D numpy.ndarray[float] the embeddings of the command phrases, one row per command,
   • 2D numpy.ndarray[float] the embeddings of the command titles, one row per command.
     The three objects shall have the same length, which equals to the number of commands provided by VSCode and installed plug-ins.
2. Matching a query. Denote the similarity measure function as $d$, the embedding of the query as $e_q$, and the embedding of a command as $e_c$, and the embedding of the command's title as $e_t$. Then the similarity of the query and the command is $max(d(e_q, e_c), d(e_q, e_t))$ -- tentatively set so. We can use other heuristics later.

[forrestbao]

To do for @parthiv2048:

1. Rename and update get_command{_titles}_using_gui.py to work with only one input file, which is default_keybindings.json
2. Update clean_commands.py and embed_commands.py according to new NLP_and_backend/readme.md
3. Update the search_commands.py, fastapi_app.py (local backend server test) and main.py (for GCP deployment)