add nextjs doc llms.txt content as resources

[huozhi]

init tool was always loading llms.txt content into the response which increased around 10-15k tokens.

After migrating them to dynamic mcp resources, it saved a lot of token.

Token Reduction:
- Before: ~17,500 tokens (init includes full llms.txt)
- After: ~1,813 tokens (init is compact)
- Reduction: ~90% fewer tokens

The process is not changed, still loading the llms.txt, but instead showing into tool response, now they're added as mcp resources