Fix NewRowSynthesis failing with category data type

[cpruijsen]

Quote string categorical values in NewRowSynthesis query filters with repr(value).

Resolves #397.

NewRowSynthesis built DataFrame.query filters by checking
real_data[field].dtype == 'O'. Pandas category and string dtypes
are not 'O', so a string value such as F was interpolated unquoted
and pandas treated it as a variable name (UndefinedVariableError).

The same metric is what DiagnosticReport uses for the Synthesis
property.

Decision

Quote with repr(value) when the categorical value is a str,
otherwise interpolate the raw value. Closed PR #458 always used
repr(row[field]). iterrows() yields numpy scalars for integer and
boolean categoricals, and on current numpy repr(np.int64(1)) is
np.int64(1), which pandas.query cannot parse, while
f'{np.int64(1)}' is 1. Can switch to always-repr after .item(),
or to passing values through local_dict.

Test plan

• test_compute_breakdown_with_category_dtype fails on main with
  UndefinedVariableError: name 'F' is not defined and passes here
• test_compute_breakdown_with_string_dtype likewise
• existing tests/unit/single_table/test_new_row_synthesis.py still
  pass (integer categoricals and multi-line strings)