feat: hide sensitive columns in PandasCompare without modifying original dfs

[sh13m]

Improves on #490. Column hashing changed to data hiding/censoring. Can now be done without affecting the original dataframes and should not blow up memory footprint.

• Realized data is already being copied internally when merging dataframes in _dataframe_merge(), with df1_unq_rows, df2_unq_rows, and intersect_rows being used downstream.
• We can modify these dataframes after _compare() is called to hide the relevant sensitive columns without modifying the original dataframes, also shouldn't introduce any significant memory pressure.
• Something not caught previously but should be fixed now is that tolerances wouldn't have worked for floats because the hashing was done prior to comparing.
• Since things are being done after the compare now instead of before, we can change the hashing to a simple column replacement of all ******* to achieve full data hiding.
  • Much faster and simpler this way (don't need to worry about nulls).
• Sensitive column hiding is now done through the hide_sensitive_columns(sensitive_columns) method, we can also reveal the hidden columns through reveal_sensitive_columns().
  • combined the setter and validation into a single private method to prevent direct attribute modification, self.sensitive_columns should only be controlled through the hide/reveal methods.
• Updated unit tests.

[fdosani]

Initial thoughts on the review. Nice work!

[sh13m]

Not sure why tests were randomly failing, might have been a github issue.

[fdosani]

Some comments and thoughts, appreciate the effort and improve on the hashing!