Triage fixes and switch to Python 3.15

[msoedov]

Fixes from the 2026-10-18 issue triage, plus the move to Python 3.15.

• LLMRefusalClassifier.is_refusal reports a "pass" judge verdict as a refusal when the reply carries punctuation, quotes or markdown #350 LLMRefusalClassifier reads the judge verdict on word boundaries, so Pass., **pass**, "pass" no longer count as refusals (which hid successful attacks). Replies with no verdict still count as refusal, as before.
• Internal Server Error In Verify Integration #149 /verify returns 400 for an unparseable spec and for any upstream 4xx/5xx, with Upstream returned HTTP <code>: <body>. Previously a bad spec or a 5xx target surfaced as 500.
• Add tests to validate dataset accessibility via Hugging Face #72 Registry test uses HF Hub metadata instead of downloading datasets (run with RUN_SLOW_TESTS=1). Marked 3 datasets inactive that now return 401: dayone3nder/jailbreak_prompt_JBB_sft_trainset, dayone3nder/general_safe_mix_jailbreak_prompt_JBB_trainset, ShawnMenz/DAN_jailbreak.
• Refusal classifier models re-saved with scikit-learn 1.9.1. 1.9 could not run the 1.6.1 pickles (missing _effective_probability). Decision scores are identical to the old files on 474 texts (max abs diff 0.0).
• poetry install fails on Python 3.14 without a C toolchain — several locked pins ship no cp314 wheel #347 / Python 3.15 requires-python >=3.15, CI, pre-commit and docs on 3.15. Full relock: every dependency installs from wheels except multiprocess and pyyaml, which build without a C compiler. Allows pandas 3 and scikit-learn 1.9. Also clears the anyio (critical), urllib3 and mkdocs-material alerts.
• Docker Dockerfile, .dockerignore and the docker build workflow removed.
• flake8 E731 ignored in .flake8.
• Fix image_generator.py:111: UserWarning: Glyph 9 ( ) missing from font(s) DejaVu Sans. plt.savefig(buffer, format="jpeg", bbox_inches="tight") #135 Already fixed in 6dec776 (with test); closing.

Closes #350, closes #149, closes #347, closes #72, closes #135

Tests: 479 passed, 5 skipped on Python 3.15.0. New tests for #350 and #149 fail on main.

🤖 Generated with Claude Code