feat: Variet Engine v1.0 + 5-model tuning complete

Phase 01 (LLM Tuning):
- Gemma4 26B: 74.65 t/s (fast)
- Qwen 35B: 61.62 t/s (balanced)
- Gemma4 31B: 16.0 t/s (deep-coder)
- Qwen 27B: 16.7 t/s (deep-logic)
- Qwen 122B: 8.95 t/s (ultra, GPU 1 only)

Phase 02 (API Engine):
- FastAPI reverse proxy on port 8000
- /engine/switch hot-swap with 503 protection
- config/engine_models.json as single source of truth
- Replaced 4 individual .bat files with unified engine

File cleanup:
- scripts/ 85 files -> 9 + _archive/
- Root .bat files -> _archive/

_archive/models_list.txt

@@ -0,0 +1,7 @@
+C:\Users\Variet-Worker\Desktop\variet-llm\models\gemma-4-26B-A4B-it-MXFP4_MOE.gguf
+C:\Users\Variet-Worker\Desktop\variet-llm\models\gemma-4-26B-A4B-it-Q4_K_M.gguf
+C:\Users\Variet-Worker\Desktop\variet-llm\models\Qwen3.5-35B-A3B-MXFP4_MOE.gguf
+C:\Users\Variet-Worker\Desktop\variet-llm\models\Qwen3.5-35B-A3B-Q4_K_M.gguf
+C:\Users\Variet-Worker\Desktop\variet-llm\models\Q4_K_M\Qwen3.5-122B-A10B-Q4_K_M-00001-of-00003.gguf
+C:\Users\Variet-Worker\Desktop\variet-llm\models\Q4_K_M\Qwen3.5-122B-A10B-Q4_K_M-00002-of-00003.gguf
+C:\Users\Variet-Worker\Desktop\variet-llm\models\Q4_K_M\Qwen3.5-122B-A10B-Q4_K_M-00003-of-00003.gguf

_archive/start_gemma4_26b_api.bat

@@ -0,0 +1,30 @@
+@echo off
+chcp 65001 >nul
+echo =========================================================
+echo  Gemma4 26B-A4B API Server (256K Context - Final Optimal)
+echo  [CORE] --n-cpu-moe 10: VRAM 12GB 최적화용 Expert 오프로드
+echo  [TUNED] -t 4 -ub 512: CPU 병목 방지 및 SWA 캐시 최적화
+echo  [PERF] Speed: ~30.9 t/s (1x RTX 3060)
+echo =========================================================
+echo.
+
+llama_bin_run\llama-server.exe ^
+  --model models\gemma-4-26B-A4B-it-Q4_K_M.gguf ^
+  -ngl 999 ^
+  --n-cpu-moe 10 ^
+  -c 262144 ^
+  -np 1 ^
+  -fa on ^
+  --cache-type-k q4_0 ^
+  --cache-type-v q4_0 ^
+  -ub 512 ^
+  -b 2048 ^
+  -t 4 ^
+  -tb 4 ^
+  --mlock ^
+  --prio 3 ^
+  --poll 50 ^
+  --port 8000 ^
+  --host 0.0.0.0
+
+pause

_archive/start_qwen_122b_api.bat

@@ -0,0 +1,10 @@
+@echo off
+echo =========================================================
+echo  Qwen3.5 122B-A10B API Server (MoE Optimized)
+echo  [INFO] --cpu-moe: Expert weights forced to CPU RAM
+echo  [INFO] -ngl 999: All attention layers on GPU
+echo  [INFO] -np 1: Single slot, -c 2048 minimal context
+echo =========================================================
+echo.
+llama_bin_run\llama-server.exe --model models\Q4_K_M\Qwen3.5-122B-A10B-Q4_K_M-00001-of-00003.gguf -ngl 999 --cpu-moe -c 2048 -np 1 -fa on --cache-type-k q4_0 --cache-type-v q4_0 -ub 256 -b 1024 -t 8 --mlock --prio 2 --port 8000 --host 0.0.0.0
+pause

_archive/start_qwen_122b_optimized.bat

@@ -0,0 +1,29 @@
+@echo off
+chcp 65001 >nul
+echo =========================================================
+echo  Qwen3.5 122B-A10B API Server (Final Optimized)
+echo  [CORE] --cpu-moe + -ngl 999: WDDM swap 완전 회피
+echo  [TUNED] -t 4: CPU 캐시 경합 최소화 (10.83 t/s)
+echo  [TUNED] mmap on: 메모리 매핑 활성 (no-mmap보다 빠름)
+echo  [VRAM] 5.4GB / 12GB (54%% 여유)
+echo =========================================================
+echo.
+
+llama_bin_run\llama-server.exe ^
+  --model models\Q4_K_M\Qwen3.5-122B-A10B-Q4_K_M-00001-of-00003.gguf ^
+  -ngl 999 ^
+  --cpu-moe ^
+  -c 2048 ^
+  -np 1 ^
+  -fa on ^
+  --cache-type-k q4_0 ^
+  --cache-type-v q4_0 ^
+  -ub 256 ^
+  -b 1024 ^
+  -t 4 ^
+  --mlock ^
+  --prio 2 ^
+  --port 8000 ^
+  --host 0.0.0.0
+
+pause

_archive/start_qwen_35b_api.bat

@@ -0,0 +1,10 @@
+@echo off
+echo =========================================================
+echo  Qwen3.5 35B-A3B API Server (MoE Optimized)
+echo  [INFO] --cpu-moe: Expert weights forced to CPU RAM
+echo  [INFO] -ngl 999: All attention layers on GPU
+echo  [INFO] -np 1: Single slot for max speed
+echo =========================================================
+echo.
+llama_bin_run\llama-server.exe --model models\Qwen3.5-35B-A3B-Q4_K_M.gguf -ngl 999 --cpu-moe -c 4096 -np 1 -fa on --cache-type-k q4_0 --cache-type-v q4_0 -ub 512 -b 2048 -t 6 --mlock --prio 2 --port 8000 --host 0.0.0.0
+pause