[BUG] Adding KPI with incorrect columns does not produce the correct errors

[kartikay-bagla]

Describe the bug

When selecting a non datetime column as the date column, we get an empty dataframe error instead of an error explaining that the datetime column is not parsable.

[kartikay-bagla]

So if we accidentally select a double/float/int column as the datetime column, pandas can infer it as unix timestamp (which may be incorrect) and technically passes our datetime columm check. But after that it leads to a SQL error.

How do we want to deal with these type of columns?

@suranah