brand: align with brand system

[amcheste-ai-agent]

Holistic pass to align this repo with the personal brand system at alanchester-brand. No code changes; markdown prose + README badges only.

Surfaces touched

• README.md
  • Title em dash → colon (Pokémon Red: Reinforcement Learning Toolkit).
  • Badge palette: License yellow → 1F4D3A (hunter-green accent per brand rule); Tests brightgreen → 0B0B0C (status decoration, demoted to ink); Python blue → 0B0B0C (version-style, ink).
  • Em-dash sweep across prose (smoke-test comment, operational-playbook parenthetical, monitoring table, license line).
• CONTRIBUTING.md — em-dash sweep across 6 prose lines.
• MODEL_CARD.md — em-dash sweep (title + 3 prose lines).
• CITATION.cff — em-dash sweep (header comment + work title).
• DEVELOPER_GUIDE.md — em-dash sweep on prose lines and API definition-list bullets; left technical body otherwise untouched.
• docs/research_playbook.md — em-dash sweep across step headers, prose, and bullet labels (~22 occurrences).
• configs/wandb_report_template.md — em-dash sweep on prose; one superlative softened (single most important figure → headline figure).
• .github/PULL_REQUEST_TEMPLATE.md, .github/ISSUE_TEMPLATE/bug_report.md, .github/ISSUE_TEMPLATE/feature_request.md — em-dash sweep on contributor-facing instruction comments.

Surfaces skipped

• CSS / SCSS / Tailwind / fonts / favicons / logos / social preview — none exist; this is a Python ML project.
• Hardcoded hex in pokemon_red_ai/analysis/comparison.py, scripts/analyze.py, pokemon_red_ai/training/callbacks.py — these are matplotlib data-viz colors used to distinguish the three RL treatments (pixel/symbolic/hybrid) in figures and live dashboards. Intentional viz convention, not theme palette. Flagged below.
• Code-block / API-docstring em dashes inside DEVELOPER_GUIDE.md body — kept conservative on technical content per the brand spec.
• Untracked docs/design/ directory — not part of this repo's tracked set; left alone.

Flags

• Repo description on GitHub still mentions "TensorBoard" but the README now describes Streamlit + W&B. Outside brand scope, but worth refreshing in a follow-up: gh repo edit --description ....
• Data-viz palette in comparison.py / analyze.py / callbacks.py is intentionally non-brand. The brand has one accent (the δ); RL paper figures need 3 distinguishable treatment colors. Suggest leaving as-is unless you want to revisit treatment colors holistically across already-rendered figures.
• The brand README's accent table lists --ac-accent: #1F4D3A while tokens/colors.css lists #1F6B3A. I followed the prompt and the README convention (1F4D3A) for the license badge; worth reconciling upstream.

Verification

• Eyeball the new README badge colors render the way you expect on GitHub.
• Re-read the swept lines in CONTRIBUTING.md and docs/research_playbook.md for natural flow (semicolons / colons / parentheses substituted for em dashes; no awkward "period + lowercase" artifacts found in the audit).
• Confirm MODEL_CARD.md and CITATION.cff title changes (em dash → colon) are fine for downstream consumers (the GitHub citation widget, BibTeX exports).
• Confirm the wandb_report_template.md voice change (single most important → headline) is acceptable.