Detect CUDA OS Overhead

This adds logic to detect skew between the driver and
management library which can be attributed to OS overhead
and records that so we can adjust subsequent management
library free VRAM updates and avoid OOM scenarios.

gpu/types.go

@@ -52,7 +52,8 @@ type CPUInfo struct {
 
 type CudaGPUInfo struct {
 	GpuInfo
-	index int //nolint:unused,nolintlint
+	OSOverhead uint64 // Memory overhead between the driver library and management library
+	index      int    //nolint:unused,nolintlint
 }
 type CudaGPUInfoList []CudaGPUInfo