docs(whitepaper): Add BENCH-001–006 complete results with analysis

- Update whitepaper to v1.0 with complete benchmark results
- Add CPU results: GF16 achieves f32 accuracy (0.00% gap) on trained MNIST MLP
- Add FPGA unit-level results: GF16 47–59× LUT overhead vs ternary (expected)
- Add FPGA MAC-level results: GF16 1.37× LUT overhead, 16× DSP blocks
- Add DSP bottleneck analysis: 15 parallel GF16 MAC-16 vs 1,219 ternary
- Add trade-off space diagrams (quality vs scalability vs energy)
- Add hybrid architecture recommendation (ternary bulk + GF16 critical)
- Add quantization analysis explaining why ternary fails catastrophically
- Update author to Dmitrii Vasilev

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: Antigravity Agent