Align Agentic Retrieval sizing for 0.23.02

[Neil Bird [Msft] (NeilBird)]

Summary

Align the Sizer's Agentic Retrieval model with the current Microsoft Foundry Local requirements and release the changes as ODIN 0.23.02.

• Model three CPU workers for every deployment and two embedding GPU workers for Combined and Knowledge modes
• Add Combined, Knowledge, and Agentic deployment modes plus external, Foundry Local minimum, and Foundry Local production LLM endpoint choices
• Allow the full AKS Arc DDA GPU list: T4, A2, A16, L4, L40, L40S, and RTX Pro 6000; present NC8_A2 and NC8_A16 as recommended sizes
• Validate local model-host VRAM requirements while continuing to reject GPUs without published AKS Arc VM SKUs
• Bump the Sizer payload to version 3 while normalizing legacy computeMode exports to a compatible Combined/external configuration
• Update JSON Schema, Markdown/HTML reports, PowerPoint output, release notes, and regression coverage
• Enforce patched js-yaml and brace-expansion floors while retaining only the exact temporary nanoid audit exception until its patched release is available

Validation

• node scripts/run-tests.js — 1,504 / 1,504 passed
• Full ESLint scope — 0 errors (2 pre-existing warnings in js/script.js)
• npx html-validate "**/*.html" — passed
• npx stylelint "**/*.css" — passed
• node scripts/check-npm-audit.js — no unapproved high or critical advisories
• node scripts/smoke-test-pptx.js — passed
• git diff --check — passed

Manual merge only after review and checks pass.