[Optimization] Cache the ranks subsample

It is probably unnecessary to always [sample](https://github.com/biocore/microsetta-public-api/blob/306e60d9ba66e0c1028e8444c557c4b96c9c5d29/microsetta_public_api/models/_taxonomy.py#L279) for ranks. Instead, we should cache once per dataset and optionally allow the client to request a new sample.

This endpoint has ~500ms turn around time on staging

Start in microsetta_public_api/models/_taxonomy.py around the sampling call at line 279 and trace the endpoint that uses it. Measure the current staging turnaround, then determine how a per-dataset cache and an optional new-sample request should behave; done means ranks no longer resample on every request while the requested refresh remains possible.

Written by the indexing model from the issue text.