[SYSTEMDS-3166] Refactor Isolation Forest builtin and add edge-case tests

[PeilinChen01]

Summary

This pull request contains the current LDE progress for SYSTEMDS-3166 and builds on the original Isolation Forest implementation in #2421.

Main changes:

• Created 10 test scenarios for two execution modes, Single Node and Hybrid, resulting in 20 automated tests.
• Added an R script for comparison.
• Implemented the transfer of a model trained by SystemDS to R.
• Verified that R and SystemDS produce the same anomaly scores when applying the same isolation forest.

Tests

The following command passes locally:

mvn -Dtest=BuiltinIsolationForestTest test

Covered test cases include:

• Basic model training and scoring
• Hybrid execution mode
• Subsampling-size clamping
• Single-row apply
• Single-tree model
• Constant data
• Seed reproducibility
• Anomaly ranking

Experiment results:

Data Size(rows)	Feature Number	Tree Number	Subsampling Size	Repetition	SystemDS(s)	R(s)	SystemDS speedup vs R
1000-64000	30	100	256	20	3.85-3.89	0.95-0.97	~0.25
1000-64000	50	100	512	3	7.56	2.55	~0.33
1000-64000	30	100	256	20	4.1-4.6	0.95	~0.22
1000-64000	3	10	20	20	0.075-0.138	0.005-0.006	~0.06

[mboehm7]

LGTM - thanks @PeilinChen01 @rbbozkurt and others for your contributions. During the merge, I only moved the java test to builtin/part1 to follow alphabetical ordering, disabled the benchmarking in the unit tests, and deleted the isolation forest from staging.