UT Coverage for cbps.py

Summary:
## Summary:
This diff improves test coverage for `fbcode/core_stats/balance/weighting_methods/cbps.py` by adding comprehensive unit tests. The changes include:

**Test Coverage Improvements:**
- Added tests for CBPS 'exact' method behavior with tight constraints
- Added tests for invalid method parameter validation
- Added tests for balance_classes=False option
- Added tests for DataFrame input handling in helper functions
- Added tests for different scipy optimization methods
- Added tests for GMM optimization result selection
- Added tests for various edge cases and error conditions
- Added tests for weight trimming with percentile method

**Code Quality Improvements:**
- Enhanced type annotations using Optional[T] instead of Union[T, None]
- Improved docstring formatting for better readability
- Fixed deprecated pandas functions (using pd.concat instead of append)
- Added DataFrame to numpy array conversion in gmm_function to handle pandas inputs correctly
- Ensured consistent handling of array and DataFrame inputs

**Test Fixes Applied:**
- Updated test_cbps_exact_method_with_constraint_violation to verify weights become identical instead of expecting exception
- Modified test_cbps_with_weight_trimming_percentile to test percentile trimming functionality correctly

Differential Revision: D89150493

fbshipit-source-id: 4e676e7943969f1e2963619a72d495cebd5d2a17

Author: talgalili

balance/weighting_methods/cbps.py

@@ -41,6 +41,9 @@ def logit_truncated(
     Given an X matrx and avector of coeeficients beta, it computes the truncated
     version of the logit function.
 
+    Truncation prevents numerical instabilities by bounding probabilities away from 0 and 1,
+    which is essential for stable weight computation in propensity score methods.
+
     Args:
         X (Union[np.ndarray, pd.DataFrame]): Covariate matrix
         beta (np.ndarray): vector of coefficients
@@ -129,6 +132,14 @@ def gmm_function(
             loss (float) computed gmm loss
             invV (np.ndarray) the weighting matrix for GMM
     """
+    # Convert inputs to numpy arrays to avoid pandas indexing issues
+    if isinstance(X, pd.DataFrame):
+        X = X.values
+    if isinstance(design_weights, (pd.Series, pd.DataFrame)):
+        design_weights = np.asarray(design_weights).flatten()
+    if isinstance(in_pop, (pd.Series, pd.DataFrame)):
+        in_pop = np.asarray(in_pop).flatten()
+
     probs = logit_truncated(X, beta)
     N = np.sum(design_weights)
     N_target = np.sum(design_weights[in_pop == 1.0])