Refactor `_select_matching_columns` to reduce nesting

[ulfaslakprecis]

ok, how about taking the shorter branch first here? I.e. if field.alias is None:

Also, what do you think about defining three small functions adding semantics and hiding the lowest level branching?

Names can be made more descriptive, but with this it would read something like:

    if field.alias is None:
      # return [column] if no error
      matched = _check_non_alias(df, column, matching_columns_in_df, optional)
      continue

    if field.regex:
        matched = cls._check_alias_regex(df, column, matching_columns_in_df, optional)
    else:
        # returns [field.alias] if no error
        matched = cls._check_alias(df, column, matching_columns_in_df, optional)

    matching_columns_in_df.extend(matched)

(Here these are class methods since cls is used in the message.)

It might also help testing to be more unit-y, i.e, testing each individual _check function.

Originally posted by @pierreprecis in #5 (comment)