feat: Phase 2 & 3 - CoreML acceleration and NLP support

[rlaope]

Summary

• Implement CoreML/Neural Engine acceleration (ort 2.0 API)
• Add NLP feature: embed command for text embeddings
• Add documentation: ARCHITECTURE.md, TUTORIAL.md, API.md

Changes

Phase 2: CoreML Acceleration

• coreml.rs: Complete rewrite with ort 2.0 API
• Add ComputeUnits enum (All, CpuAndNeuralEngine, CpuAndGpu, CpuOnly)
• bench.rs: Add neural-engine provider option

Phase 3: NLP Support

• Add embed CLI command
• Add run_multiple() method for multi-input models
• Text tokenization and embedding generation

Documentation

• docs/ARCHITECTURE.md: Internal architecture explanation
• docs/TUTORIAL.md: Step-by-step tutorials
• docs/API.md: Complete API reference
• README.md: Updated with new features

Test plan

• cargo build --release --features coreml,nlp
• Test airml embed command
• CoreML benchmark comparison

🤖 Generated with Claude Code