`NaiveHinter` - Add thresholds calibrator.

Find a few hint examples.
Automate a process where the model is grading these examples, and the thresholds parameters are determined based on the distances the model give to the examples.

Start by locating NaiveHinter, its threshold parameters, and the code that grades hint examples or computes model distances. Gather representative hint examples, then define how calibration should derive thresholds from those distances. Done means the process runs automatically and produces usable thresholds for NaiveHinter.

Written by the indexing model from the issue text.