Fix model pricing: correct $0 prices for GPT-5.4 and Claude Sonnet 4.6

[rajbos]

Problem

Cost estimates were significantly underreporting because two of the most heavily-used models had $0.00 pricing, and two reasoning models were overpriced by 3.6×. The pricing data sources hadn't been refreshed since January 16, 2026.

Critical Fixes

Model	Before (in/out per M)	After (in/out per M)	Impact
gpt-5.4	$0.00 / $0.00	$2.50 / $15.00	🔴 All GPT-5.4 usage was FREE
claude-sonnet-4.6	$0.00 / $0.00	$3.00 / $15.00	🔴 All Sonnet 4.6 usage was FREE
o3-mini	$4.00 / $16.00	$1.10 / $4.40	Was 3.6× too high
o4-mini	$4.00 / $16.00	$1.10 / $4.40	Was 3.6× too high

New Models Added

Model	Input	Output	Tier
gpt-5.4-mini	$0.25	$2.00	standard
gemini-3.1-flash-lite	$0.25	$1.50	unknown

New Pricing Sources

Added 3 new sources for better accuracy and verification going forward:

• xAI Grok API (https://x.ai/api) — covers Grok models
• GitHub Copilot Premium Requests — validates multiplier values
• OpenRouter (https://openrouter.ai) — cross-provider pricing aggregator for verification

Also added the official Anthropic platform docs URL alongside the marketing page.

Files Changed

• vscode-extension/src/modelPricing.json — pricing corrections, new models, new sources
• vscode-extension/src/tokenEstimators.json — token ratios for new models
• vscode-extension/src/README.md — updated source URLs and Gemini pricing section
• .github/skills/refresh-json-data/SKILL.md — updated source list

Verification

• ✅ npm run compile passes (extension)
• ✅ npm run build passes (CLI)
• ✅ JSON validation passes (50 models, 50 estimators, 7 sources)