Automatically update training data from source collector

[josh-chamberlain]

Context

We're starting to generate more sources, but we don't update our training data.

Requirements

• We need a script which we can automate to update our training data—manually for now, just before we retrain (which is infrequent).
• check for newly labeled stuff from our source collector app
  • We want to grab not only items labeled as "relevant" and in our db, but also items labeled not relevant.
• Collect HTML responses upstream of annotation #324
• optional: apply keyword extraction to labeled data
• Submit approved URL Task #152
• update the hugging face training-urls dataset, with batch ID
  • Tweak Data Collection #141
  • place it there as raw data; we can transform it into more specific datasets as needed

[maxachis]

Since we're no longer using Label Studio, I think this can be safely closed.

[josh-chamberlain]

@maxachis I abstracted it—we don't use label studio, but our own labeling app should still cause better training data! we're a ways from having enough sources, though, so it's a future thing