Isolation Forest follow-up work #8420

Tracks deferred robustness, scalability, interoperability, API-parity, and cuml.accel work following #8226.

C++ API robustness

• Handle n_rows == 0 in the public C++ scoring path before launching a zero-block CUDA kernel, and use checked launch-dimension casts. Isolation Forest Implementation #8226 (comment)

Sampling quality and scalability

• Avoid the linear duplicate checks used for row and feature sampling at large max_samples or max_features. Consider a fast path when max_samples == n_rows and a scalable sampling strategy. Isolation Forest Implementation #8226 (comment)
• Select uniformly among nonconstant features when splitting a node, avoiding sensitivity to column order. Isolation Forest Implementation #8226 (comment)

Python interoperability and persistence

• Support fitted IsolationForest conversion to scikit-learn #8479
• Remove the local workaround for missing n_node_samples in exported isolation forest #8488
• Investigate using nvforest for inference, and dropping the C++ model entirely (++)
• Preserve fitted model behavior across pickle round trips, potentially using the serialized Treelite model and nvForest inference path, and remove the check_estimators_pickle xfail. Remove native model from Isolation Forest estimators #8493
• Raise NotFittedError from unfitted estimator methods and remove the corresponding common-estimator-check xfail. Raise NotFittedError from unfitted IsolationForest methods #8475
• Implement or deprecate the unsupported sample_weight argument FIX Remove spurious IsolationForest arguments #8486
• Implement or deprecate the unsupported warm_start argument FIX Remove spurious IsolationForest arguments #8486
• Support estimator_, estimators_, estimators_features_, and estimators_samples_ attributes. Partly done in Support fitted IsolationForest conversion to scikit-learn #8483 for cuml.accel. estimators_samples_ is not done yet there.

cuml.accel support

• [FEA] Add cuml.accel support for IsolationForest #8468

Treelite usage

• Set task_type=kIsolationForest in the Treelite representation
• Use the built-in postprocessor exponential_standard_ratio in Treelite/nvForest; remove the transform s(x) = 2^(-E[h(x)] / c(n)) from the Python layer.
• Set data_count and data_count_present. Right now, cuML IsolationForests sets all data_count as 0.

Follow-up to #8226.
Related to #6096.