fix: restore optimal v3 models + add fitness gate protection

- Restore all 30 agents to v3.html heatmap optimal models:
  * frontend-developer: qwen3-coder -> minimax-m2.5 (92★)
  * devops-engineer: nemotron-3-super -> kimi-k2.6:cloud (88★)
  * browser-automation: qwen3-coder -> kimi-k2.6:cloud (86★)
  * agent-architect: glm-5.1 -> kimi-k2.6:cloud (86★)
- Add Model Evolution Guard system:
  * agent-evolution/scripts/lib/fitness-gate.cjs
  * Rejects downgrades >3 points or below score 75
  * Produces detailed diff report before any file modifications
  * Normalized model ID lookup (v3.html ':' vs JSON '-')
- Update sync-benchmarks-from-yaml.cjs with fitness gate
- Update model-benchmarks.json with v3 optimal assignments
- Rebuild research-dashboard.html (104KB, 30 agents, 11 models)
- Add model-evolution-guard.md architecture documentation
- Add v3-optimal-models.json as source-of-truth reference

Fixes regression introduced by commit 3badb25 where models were
silently downgraded from heatmap optimal to inferior assignments.

kilo.jsonc

@@ -45,7 +45,7 @@
     "system-analyst": {
       "description": "Designs technical specifications, data schemas, and API contracts before implementation",
       "mode": "subagent",
-      "model": "ollama-cloud/nemotron-3-super"
+      "model": "qwen/qwen3.6-plus:free"
     },
     "sdet-engineer": {
       "description": "Writes tests following TDD methodology. Tests MUST fail initially (Red phase)",
@@ -68,7 +68,7 @@
     "lead-developer": {
       "description": "Primary code writer for backend and core logic. Writes implementation to pass tests",
       "mode": "subagent",
-      "model": "ollama-cloud/nemotron-3-super",
+      "model": "ollama-cloud/qwen3-coder:480b",
       "color": "#DC2626",
       "permission": {
         "read": "allow",