fix(steps): add unit test and fix null cols for impute

[jitingxu1]

changes:

1. add unit test for _impute.py
2. raise a value error if one column is all NULL, because it does not fulfill users' original goal.

resolve #117

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.35%. Comparing base (6582682) to head (5e78804).
Report is 18 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #157      +/-   ##
==========================================
+ Coverage   85.39%   89.35%   +3.95%     
==========================================
  Files          26       28       +2     
  Lines        1924     2047     +123     
==========================================
+ Hits         1643     1829     +186     
+ Misses        281      218      -63     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[deepyaman]

Left some comments, but maybe more fundamentally—why is this necessary/what motivated it? If somebody wants to fill their missing values with None, should I care? E.g. maybe they want to represent all the missing value as NULL (and not NaN), so they do this. Alternatively if you see scikit-learn SimpleImputer, they later let the user customize what is considered "missing".

[deepyaman]

Left some comments, but maybe more fundamentally—why is this necessary/what motivated it? If somebody wants to fill their missing values with None, should I care? E.g. maybe they want to represent all the missing value as NULL (and not NaN), so they do this. Alternatively if you see scikit-learn SimpleImputer, they later let the user customize what is considered "missing".

I'm pushing to deprioritize this beyond today, unless you feel it's particularly pressing. Let's focus on getting the other PRs in for release.

[jitingxu1]

Left some comments, but maybe more fundamentally—why is this necessary/what motivated it? If somebody wants to fill their missing values with None, should I care? E.g. maybe they want to represent all the missing value as NULL (and not NaN), so they do this. Alternatively if you see scikit-learn SimpleImputer, they later let the user customize what is considered "missing".

Instead of raise an error, I change it to throw a warning to the user. User should know the imputing does not work as expected, since there are still some nulls in the column

Why not?