Port call density notebook to Hoplite

[sdenton4]

The previous analysis notebook includes a useful workflow for validation and call density estimation:
https://github.com/google-research/perch/blob/main/analysis.ipynb
paired with library code here:
https://github.com/google-research/perch/blob/main/chirp/inference/call_density.py

We should reintroduce a validation notebook for Hoplite.

[sdenton4]

Discussed the following action plan:

• Produce a QuantileValidationDataManager (perhaps directly re-using the DataManager interface from the classifier training) to produce batches of validation data for human annotation.
• Re-use existing components to display + annotate those validation examples, very similar to the agile.py noteboook.
• Store validation results as labels in the database.
• Use the validation labels to compute call density.

We observe that the ValidationDataManager samples windows w with some probability P(w). This P(w) needs to be stored along with the validation label to allow accurate computation of call density. This adds 1-3 action points to the plan:

• Store the ValidationDataManager sampling probability in the Labels table alongside the validation decision.