docs(research): Add comprehensive KFuncs research for AI inference in eBPF

Added 8,000-word research document covering KFuncs as the modern extensibility
mechanism for eBPF programs, with focus on ML inference capabilities.

Key sections:
- KFuncs vs Helper Functions (why KFuncs are the future)
- Proposed ML KFuncs (matmul, activations, hardware acceleration)
- Complete kernel module implementation with BTF type safety
- SBPFuncs proposal for Solana runtime extensibility
- Governance and security considerations

This completes the trilogy of eBPF/AI research documents supporting the
OVSM textbook's on-chain execution chapter.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>

Author: 0xrinegade