Augplot 0.3.0

Added

• ChatGPT sign-in with secure credential storage, account switching, model discovery,
  and per-account model selection, alongside existing API-provider configuration.
• Full-input category counts and saved rejected-generation diagnostics, with an optional
  second repair attempt through max_repairs=2.
• Reusable count-weighted empirical smoothing and renderer-aware direct curve labels,
  with SciPy-backed helpers bundled into standalone Python exports.
• Bounded grouped relationship profiles, numeric quantiles, and explicit coverage,
  sampling, log-transform, and truncation metadata.
• Optional max_model_calls budget shared across generation, repairs, and clarification,
  and inspectable attempted-call counts on charts and Augplot failures.
• An opt-in H–R visual-evaluation harness and unrelated complex-plot regressions.

Changed

• Accept single-binding pyplot colormap calls and bounded literal numeric formatting,
  including Unicode-aware format-width limits.
• Guide generation toward statistical smoothing and consistent external-colorbar layout.
• Version helper/profile changes in history identity while preserving existing workflow,
  live Figures, provider setup, and execution guardrails.
• Add SciPy 1.11.4 or newer as a runtime dependency for shared smoothing.
• Refresh the H–R notebook with a short appearance-focused request and strict fresh
  one-call generation. Document sparse and contaminated class-label limits.

Fixed

• Allow library-internal date-formatting imports and NumPy dtype=object annotations
  while retaining generated-code import and reflection restrictions.
• Report undefined generated names in repair diagnostics without exposing raw data or
  library-internal names, and mask account identity in ChatGPT plotting progress.
• Avoid duplicate inline figures in the H–R notebook's standalone-export example.