Validate NVML_FI_DEV_GET_GPU_RECOVERY_ACTION gpu state during health check

It is appeared that DCGM is not able detect GPU which is stuck in EBUSY state because of pended recovery aciton

Example:

# GPU=1 stuck in EBUSY state because some GSP issue
nvidia-smi --query-gpu=index,gpu_recovery_action --format=csv,noheader
0, None
1, Reset   <= GPU is unusable, reset required 
2, None
3, None
4, None
5, None
6, None
7, None

dcgmi dmon  -c 1 -e 230
#Entity    XIDER        
ID                      
GPU 0      0            
GPU 1      120          
GPU 2      0            
GPU 3      0            
GPU 4      0            
GPU 5      0            
GPU 6      0            
GPU 7      0   

This means that GPU is completely unusable, but health check report nothing

$ dcgmi  health -s a
Health monitor systems set successfully.
$ dcgmi  health -c -j
{
        "body" : 
        {
                "Overall Health" : 
                {
                        "value" : "Healthy"
                }
        },
        "header" : 
        [
                "Health Monitor Report"
        ]
}

Start by tracing the health-check handling for NVML_FI_DEV_GET_GPU_RECOVERY_ACTION and compare it with the dcgmi health -c -j behavior shown in the report. Reproduce the EBUSY/Reset state with the supplied nvidia-smi and dcgmi commands; done means the health report identifies the GPU as unhealthy when recovery action is Reset.