Clarify DifferenceInDifferences.predict() support contract

[shawcharles]

Problem

DifferenceInDifferences.predict() exists on the estimator, but it raises NotImplementedError. The method docstring currently describes prediction as if it were supported, while TODO.md lists predict() as a known limitation.

That can confuse users coming from sklearn-style estimators: the method is discoverable, but the supported post-estimation surface is actually results_.fitted_values for fitted training-data predictions.

Proposed behaviour

Clarify the current contract without implementing new prediction semantics:

• document that out-of-sample predict() is currently unsupported;
• keep the method raising NotImplementedError after fitting;
• point users to results_.fitted_values for fitted training-data predictions;
• frame broader prediction support as pending a post-estimation result-object design.

Proposed scope

• Update the DifferenceInDifferences.predict() docstring and error message.
• Add a regression test for the unsupported fitted-predict contract.
• Add a short API docs note under DifferenceInDifferences.
• Do not implement prediction or change result-object semantics.

Validation

• PYTHONPATH=. DIFF_DIFF_BACKEND=python pytest tests/test_methodology_did.py::TestResultsObject::test_predict_contract_points_to_fitted_values -q
• PYTHONPATH=. DIFF_DIFF_BACKEND=python pytest tests/test_methodology_did.py::TestResultsObject -q
• PYTHONPATH=. DIFF_DIFF_BACKEND=python pytest tests/test_methodology_did.py -q
• PYTHONPATH=. DIFF_DIFF_BACKEND=python .venv-py39/bin/python -m pytest tests/test_methodology_did.py::TestResultsObject::test_predict_contract_points_to_fitted_values -q
• python -m ruff check diff_diff/estimators.py
• git diff --check

Note: make -C docs html SPHINXOPTS="-W" is currently blocked on pre-existing tutorial 19 DID unknown-target warnings, which are addressed separately in PR #572.