Fill BERs with archetypes without exceeding resources

[rdmolony]

Is your feature request related to a problem? Please describe. Filling census buildings with unknown BERs exceeded streamlit sharing's resource limit

[rdmolony]

In commit 76bc62943d4d8e3cdc3456a635579d1eef2238de archetyping was removed to avoid this. Profiling of memory usage shows that as of 19/07/2021 that the 1GB RAM limit is far exceeded.

Possible solutions:

• Convert this app to a GUI so that users can use their local resources instead of streamlit free-tier - https://discuss.streamlit.io/t/using-pyinstaller-or-similar-to-create-an-executable/902
• Replace pandas with modin or pypolars to load data lazily
• Replace pandas with dask and set the worker resource limit to 1GB
• (Temporary) Flag if running locally or on cloud and only run archetyping if local