Remove conformal benchmark gist (#57)

The conformal walk-forward benchmark gave NNS.ARMA.optim access to each
evaluation chunk during model selection and interval calibration, so its
published results were not a valid out-of-sample comparison. Rather than
maintain a corrected benchmark, remove the gist entirely.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TX6vFofX1vnanE147tJFWF

Author: claude