Use sampling seed to standardize records subsample in test_pufcsv.py

[martinholmer]

This pull request eliminates randomness in the sub sample selected in the sampling test added in pull request #844 (as fixed in #864). It also does two other things. First, it moves the comparison of combined tax liabilities generated by the sub-sample and the full-sample into the test_agg() function in order to reduce test execution time. Second, it uses a sampling random-number seed that produces a relatively small difference under current-law policy between the sub-sample and full-sample combined tax liability. Below is the maximum (among the ten years of results from 2013 to 2022) relative difference between the sub-sample combined tax liability and the full-sample combined tax liability for each of nine tested sampling random-number seeds:

SEED  MAX RELATIVE DIFFERENCE (%)
---------------------------------
  10  -1.66 %
  20  -2.65
  30  +3.18
  40  -5.90
  50  +2.20
  60  -2.77
  70  +3.81
  80  +0.61
  90  -2.41
---------------------------------

These results show that picking the "right" sampling seed will make a big difference in user satisfaction.

@MattHJensen @feenberg @talumbau @Amy-Xu @GoFroggyRun @zrisher @codykallen

[codecov-io]

Current coverage is 98.12% (diff: 100%)

Merging #869 into master will increase coverage by <.01%

@@             master       #869   diff @@
==========================================
  Files            13         13          
  Lines          1816       1818     +2   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits           1782       1784     +2   
  Misses           34         34          
  Partials          0          0          

Powered by Codecov. Last update 9772cd1...430e82f