feat(research): add comprehensive codebase synthesis and formal proofs

- CODEBASE_LITERATURE_SYNTHESIS_V1.md (~800 LOC)
  * Deep architecture analysis of sacred attention
  * 2024-2026 literature comparison (BitNet, TeLLMe, TerEffic, LUT-LLM)
  * Mathematical foundation derivation (Trinity Identity)
  * Gap analysis with SOTA (3 critical, 2 medium, 2 low)
  * 6-month research roadmap

- FORMAL_PROOFS_TRINITY_V1.md (~700 LOC)
  * Theorem 1: Trinity Identity (φ² + φ⁻² = 3)
  * Theorem 2: Sacred Scale Bounds (3× ratio for d∈[64,128])
  * Theorem 3: Adaptive Scale Convergence
  * Theorem 4: Monotonic Decrease
  * Theorem 5: Gradient Amplification (3.2×)
  * Theorem 6: Ternary Additive Property (no multiplication)
  * Theorem 7: φ-RoPE Position Invariance

- EXPERIMENTAL_DESIGN_TEMPLATES_V1.md (~650 LOC)
  * YAML experiment template with statistical design
  * JSONL training log schema
  * Statistical analysis scripts (Python)
  * Power analysis calculator
  * Figure templates (training curves, box plots)
  * LaTeX table generation
  * Reproducibility checklists

All documents English-only, scientifically detailed, publication-ready.

Related: #415

Author: Antigravity Agent