Add html5ever architecture documentation

Documents why html5ever is 4x faster than Swift:
- Tendril: 16-byte string with 8-byte inline storage (zero alloc for ~80% of tokens)
- SIMD: SSE2/NEON scanning 16 bytes at a time in tokenizer hot path
- SmallCharSet: 64-bit bitmap for O(1) character classification
- BufferQueue: Zero-copy chunking between special characters
- Macros: Compile-time state machine code generation
- Rc vs ARC: Non-atomic reference counting
- Text node merging: Coalesces adjacent text

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

Co-Authored-By: Claude Opus 4.5 <[email protected]>

Author: kylehowells