feat: add does_not_start_with and does_not_end_with operators

[Rdag15]

Summary

Adds two new string operators, does_not_start_with and does_not_end_with, the negated complements of the existing starts_with / ends_with operators.

Closes #1153.

Implementation

The operators follow the negation pattern already established in dataframe_operators.py (e.g. does_not_contain is ~self.contains(...) and does_not_equal_string_part is ~self.equals_string_part(...)):

@log_operator_execution
@type_operator(FIELD_DATAFRAME)
def does_not_start_with(self, other_value):
    return ~self.starts_with(other_value)

@log_operator_execution
@type_operator(FIELD_DATAFRAME)
def does_not_end_with(self, other_value):
    return ~self.ends_with(other_value)

Changes

• cdisc_rules_engine/check_operators/dataframe_operators.py — the two operator methods
• resources/schema/rule/Operator.json — operator enum entries
• resources/schema/rule/Operator.md — operator documentation
• resources/schema/rule-merged/Operator.json — regenerated with python scripts/merge_schema_markdown.py followed by npx prettier ... --write (not hand-edited), so it stays in sync with check-schema-markdown
• tests/unit/test_check_operators/test_string_comparison.py — parametrized tests for both operators across PandasDataset and DaskDataset

Testing

• pytest tests/unit/test_check_operators/test_string_comparison.py — 77 passed (incl. 4 new cases)
• black --check — clean
• flake8 — clean
• merged schema regenerated via the project script, so check-schema-markdown stays green

[RamilCDISC]

Starts_with and ends_with currently returns raw result of pandas string methods. These methods can produce nullable/objects results when target values is missing. This can mix badly with use of ''. Other operators safely use '' when positive has already normalized its output to boolean series. Please make starts_with and ends_with null safe.

[Rdag15]

Fixed in f865a42. Confirmed: with a missing target value, starts_with returned [True, None, False] (object dtype) and does_not_start_with then raised TypeError: bad operand type for unary ~: 'NoneType'. Both starts_with and ends_with now pass na=False to str.startswith/str.endswith, so the result is always a clean boolean Series (missing -> False) that negates safely. Verified on PandasDataset and DaskDataset.

[RamilCDISC]

Could you please add cases for when 'value_is_literal=True'?

[Rdag15]

Added in f865a42: test_starts_with_literal and test_ends_with_literal cover value_is_literal=True for both operators and their negations, plus test_does_not_start_with_is_null_safe / test_does_not_end_with_is_null_safe for the missing-value case — all parametrized across PandasDataset and DaskDataset.

[RamilCDISC]

The PR adds does_not_start and does_not_end with operators to CORE. These operators are negatives of starts_with and ends_with operators. The PR was validated by:

1. Reviewing the PR for any unwanted code or comments.
2. Reviewing the PR in accordance to the AC.
3. Reviewing the code to verify it follows the previous implemented code standards.
4. Ensuring the documentation is updated.
5. Ensuring the documentation mentions the use and explanation correctly.
6. Ensuring the schema is updated.
7. Ensuring previous tests are updated and new tests are added for the new operators.
8. Ensuring the new tests cover null and empty cases.
9. Ensuring the existing operators are updated to work with '~' in the negative operators.
10. Ensuring all unit and regression tests pass.

[gerrycampion]

Am I misunderstanding, or shouldn't these be [False, True, True]