docs: advanced-AutoML sample + full doc/PDF coverage of the new features

[ancongui]

What & why

The latest features — explainability, calibration, stacking ensembles, PR-AUC selection, CV strategies, persisted audit trail — shipped with reference docs but were missing from the samples, the landing pages, and the PDF. This PR closes those gaps so every feature is demonstrated, discoverable, and in the guide.

New runnable sample (test-covered, real data)

samples/advanced_automl.py — one cohesive sample on the real breast_cancer dataset that exercises all the new capabilities:

• StratifiedKFold cross-validation strategy
• PR-AUC selection (metric="average_precision")
• a calibrated stacking ensemble (calibrate=True, ensemble=True) → Brier score
• global explainability (result.explain(...))
• a persisted JSONL audit trail of every GenAI gate decision

It auto-detects an LLM key: deterministic StaticFeatureProposer offline, real AgentFeatureProposer when a key is present — so the same sample runs in CI and against a live model.

Tested for real (incl. a live LLM)

• tests/samples/test_advanced_automl.py: offline smoke test (CI) + an @integration test against a real LLM.
• All 3 real-LLM integration tests pass with a live Anthropic key (this sample + the existing showcase: feature engineering & agentic loop). The credential was used only as a transient, out-of-repo env var and never written to the repo or any output.

Docs + PDF

• README.md, docs/README.md: Explainability added to the doc tables; AutoML rows note calibration/ensembling/PR-AUC/CV.
• docs/index.md: enriched AutoML card + a new "Explainable & trustworthy" pillar.
• docs/samples.md: five samples now, with the advanced sample documented (offline + real-LLM run lines).
• docs/brief/firefly-datascience-complete-guide.pdf: regenerated (62pp) — Ch.8 gains calibration/ensembling/PR-AUC/CV + explainability; Ch.14 the hands-on advanced selection; Ch.15 the persisted audit trail; glossary terms added. Rendered pages were visually verified.

Gates

ruff format --check + ruff check clean; no src changes this PR (docs/sample/test/PDF only), so existing suites are unaffected; the new offline sample test passes.