NLP Caching

We want to cache results from the NLP so data is not run multiple times on the same text. Additionally this should improve performance in cases where multiple dashboards for the same text are open.

We may want to run the full set of indicators on the student text every so often to always try and provide information immediately to teachers, even if the indicators are a little bit out of data.

This could include indexing by text or by document id.

The cache should include which indicators were run, the results from the indicators, and additional meta data about the cache (like process id).

No files, tests, or entry points are named. Start by tracing where NLP indicators run and how dashboard requests reach that work, then define the cache key and freshness behavior from the issue. Done means repeated analysis of the same text avoids duplicate work while retaining indicator results and cache metadata, with performance and freshness behavior verified.

Written by the indexing model from the issue text.