docs: comprehensive README and CLAUDE.md refresh for 113-crate state

Rewrites README from scratch to accurately reflect:
- 113 crates (was "60+"), 3,500+ tests (was "2,100+")
- GPU Perf Batches 1–19 complete (Blackwell SM_120)
- Full optimizer/loss/training technique inventory
- Inference stack with speculative decoding options
- Distributed stack with FSDP2/TP/PP/CP/elastic
- Vision architecture zoo, specialized domain stacks
- JEPA platform section: existing building blocks + roadmap
- Accurate CLI, benchmarks, and quick-start examples

Updates CLAUDE.md tagline to reflect current goals and batch count.

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

CLAUDE.md

@@ -1,6 +1,6 @@
 # rustytorch
 
-> GPU-accelerated ML framework in pure Rust — full PyTorch-equivalent with CUDA/Metal/ROCm/WebGPU backends, Flash Attention, Mixture of Experts, speculative decoding, federated learning, and domain-specific stacks for medical imaging, neuroimaging, and scientific computing.
+> GPU-accelerated ML framework in pure Rust — full PyTorch-equivalent with CUDA/Metal/ROCm/WebGPU backends, 19 rounds of Blackwell SM_120 optimizations, complete transformer training arsenal (Muon/Shampoo/SOAP/ScheduleFree/DPO/TIES-DARE/MoD), vLLM-class inference (Medusa/EAGLE/Lookahead speculative decoding, paged KV cache, chunked prefill), federated learning, and domain-specific stacks for medical imaging, neuroimaging, and scientific computing. Primary goal: premier JEPA self-supervised learning platform for multi-node cluster.
 
 ## Problems It Solves
 
@@ -12,7 +12,7 @@
 
 ## What It's Comprised Of
 
-108 crates organized across 8 layers:
+113 crates organized across 8 layers (GPU Perf Batches 1–19 complete as of 2026-06-27):
 
 ### Layer 1: Core Infrastructure (26 crates)
 | Crate | Role |