Mixed execute_on for PPs and scalars in CSV leads to bad output

[joshuahansel]

Bug Description

When a simulation has both scalar variables and PPs and uses CSV output and the values for execute_scalars_on and execute_postprocessors_on are different, then the columns in the output file get messed up.

Steps to Reproduce

Run the following input file:
csv.txt
And observe the CSV output:

time,aux0_0,aux0_1,aux1,aux2
0,12,13,5,10
1,12,13,5,10
2,0,0,0,0,3,2

In the header row and 2 following rows, there are 5 columns (time plus 4 scalar variables). In the last row, there are 7 columns.

Impact

Incorrect output files, which can lead to incorrect interpretation of results.

[joshuahansel]

I'm guessing it's the same for reporters. I'm going to try throwing an error if any are different and see if any apps trigger it.