Implement UP_STOP_ON_FAIL option to manage GPU state on installation failure

- Added a new configuration option `UP_STOP_ON_FAIL` to control whether the GPU should be stopped automatically if the `up` command fails, enhancing user control over resource management.
- Updated the CLI to include a `--keep-on-fail` flag, allowing users to prevent GPU shutdown during installation errors.
- Enhanced the installation scripts and documentation to reflect these changes, providing clearer guidance on the new behavior and configuration options.
- Improved error handling in the CLI to ensure proper cleanup of resources in case of failure, preventing unexpected billing for unused GPU resources.

gpu-rent.vars.example

@@ -22,6 +22,7 @@
 # IDLE_MINUTES=60
 # IDLE_GRACE_MINUTES=90
 # NOTIFY_READY=true
+# UP_STOP_ON_FAIL=true   # при падении up гасить GPU (default); false / --keep-on-fail
 # Prefer auto /32 (leave unset). Spike/WARP only:
 # GPU_RENT_SSH_CIDR=0.0.0.0/0
 
@@ -30,23 +31,27 @@
 # Подробнее: docs/llm.md
 # ---------------------------------------------------------------------------
 
-# --- A) Быстрый default: SwarmUI + llama.cpp, Ubuntu Vulkan prebuilt ---
+# --- A) Default: SwarmUI + llama.cpp (auto: CUDA если nvcc на VM, иначе Vulkan) ---
 # LLM_RUNTIME=llamacpp
 # ENABLE_SWARMUI=true
 # LLAMACPP_TAG=b10545
-# (GGUF — llamacpp-models.yaml; бинарь без compile)
+# LLAMACPP_BACKEND=auto
 
-# --- B) Только LLM, без SwarmUI (дешевле по времени bootstrap / VRAM) ---
+# --- B) Только LLM, без SwarmUI ---
 # WORKLOAD=llm
 # LLM_RUNTIME=llamacpp
 # LLAMACPP_TAG=b10545
 
-# --- C) Максимальная скорость инференса: CUDA-сборка (5–15 мин первый раз) ---
+# --- C) Явно CUDA (и снести старый Vulkan-бинарь) ---
 # LLM_RUNTIME=llamacpp
 # LLAMACPP_TAG=b10545
-# LLAMACPP_BUILD_CUDA=1
+# LLAMACPP_BACKEND=cuda
 # LLAMACPP_FORCE_REINSTALL=1
-# (FORCE — снести старый Vulkan/битый бинарь и переустановить)
+
+# --- C2) Явно быстрый Vulkan, без compile ---
+# LLM_RUNTIME=llamacpp
+# LLAMACPP_BACKEND=vulkan
+# LLAMACPP_BUILD_CUDA=0
 
 # --- D) Свой бинарь / pin URL (supply-chain) ---
 # LLM_RUNTIME=llamacpp