Banker's Wrapped v1.9.0

Maintenance + release-hardening pass ahead of submission. No runtime behavior change.

Changed

• Dependencies refreshed (uv lock --upgrade): uvicorn 0.51, fastapi 0.139, openai 2.45,
  boto3 1.43.46, mypy 2.2, plus patch-level transitives. No major bumps; the Genblaze
  subpackages (genblaze-core 0.3.4, genblaze-s3 0.3.4, genblaze-gmicloud 0.3.2) were
  already the newest published — the v0.4.0 release wave.
• Coverage 97.8% → 99.6% (134 → 148 tests): analytics achievements + ACHIEVER path,
  invalid-category fallback, narrative prompt/NIM/0-scene branches, and recap
  413 / pipeline-failure / malformed-insights / empty-artifacts / progress-callback paths.
• Coverage claim synced 98% → 99% across README, CLAUDE.md, and submission docs;
  regenerated vo_08-production + vo_full-reference demo voiceover clips to match.

Fixed

• config.openai_model default was a stale gpt-4o — it is a fallback models_used.llm
  label only, now set to the real chat model and documented as such.
• Clarified that config.gmi_chat_model defaults to NIM's free dev tier on purpose
  (prod overrides via GMI_CHAT_MODEL); added an ADR-011 cross-reference in ADR-003.
• Accuracy: corrected Semantic Kernel references (README badge/diagram/tech-stack,
  ADR-002, base.py docstring, scaffold changelog line) — it was never a dependency or
  used; agents are a hand-rolled typed async BaseAgent pipeline, now documented as such.

---