Precip + no-precip input time steps cause blending failure AND in general ar_order=1 causes blending failure

[FelixE91]

We noticed that https://github.com/pySTEPS/pysteps/blob/master/pysteps/blending/steps.py fails if your radar input consists of time steps with precipitation and time steps without precip. That is because we cannot calculate the correlation coefficient with 0 variance.

While trying to fix this issues, we also noticed that the blending fails for ar_order=1. This is a more general issue.
There are no tests for ar_order=1, too.

[FelixE91]

Little addition to what we hope to fix:
# ### ... loaded the data-based precip array before
zero_value = np.nanmin(precip) #get data-based zero precip value
fullprecip = np.random.random((3,700,700))*4-2 #set the artificial radar input to test the cases
# # Case 1: All precip steps
# precip = fullprecip.copy()
# # Case 2: All no precip steps
# precip = np.ones(fullprecip.shape)*zero_value
# # Case 3: Latest no precip
# precip = fullprecip.copy()
# precip[-1] = zero_value
# # Case 4: Oldest no precip
# precip = fullprecip.copy()
# precip[0] = zero_value
# # Case 5: Latest with precip
# precip = np.ones(fullprecip.shape)*zero_value
# precip[-1] = fullprecip[-1]
# # Case 6: Oldest with precip
# precip = np.ones(fullprecip.shape)*zero_value
# precip[0] = fullprecip[0]
# # # Case 7: Precip step between 2 no precip steps
# precip = np.ones(fullprecip.shape)*zero_value
# precip[1] = fullprecip[1]
# # Case 8: No precip step between 2 steps with precip
# precip = fullprecip.copy()
# precip[1] = zero_value

### Currents steps.py: only case 1 and case 2 succeed

### Modified steps.py: case 1, case 2, case 3, case 6, case 7 succeed
###      -> I hope that we can have cases 1,2,3,4,6,7,8 to succeed with ar_order=1 fix
###         only case 5 will fail and it will be caught as error message in check_previous_radar_obs()