bench: limit quick mode to 2k groups, add performance warnings

Changes:
- Reduce 'quick' mode from 5k to 2k groups (robust is too slow)
- Add warning comment about robust performance on small groups
- Document that robust is 85-90× slower than v2 on small groups
- Document that v4 is 17,000-40,000× faster than robust

Findings from benchmarks:
- Robust: ~26 s/1k groups (small groups)
- v2:     ~0.3 s/1k groups
- v4:     ~0.0001 s/1k groups

Recommendation: Use v2/v4 for small groups (< 50 rows/group).
Robust is designed for large groups with robust statistics needs.

Note: Large numerical disagreement (0.57) observed on small groups.
This requires investigation but is deferred post-restructuring.

Author: miranov25

UTILS/dfextensions/groupby_regression/benchmark_results/history.csv

@@ -0,0 +1,10 @@
+date,host,commit,scenario,engine,n_groups,rows_per_group,duration_s,per_1k_s,speedup,notes
+2025-10-25T15:45:17.308904,Marians-MBP-3.fritz.box,unknown,Tiny (100×5),robust,100,5,2.680,26.797,1.00,
+2025-10-25T15:45:17.308911,Marians-MBP-3.fritz.box,unknown,Tiny (100×5),v2,100,5,0.778,7.777,3.45,
+2025-10-25T15:45:17.308916,Marians-MBP-3.fritz.box,unknown,Tiny (100×5),v4,100,5,0.684,6.840,3.92,
+2025-10-25T15:45:17.308919,Marians-MBP-3.fritz.box,unknown,Small (1k×5),robust,1000,5,25.852,25.852,1.00,
+2025-10-25T15:45:17.308923,Marians-MBP-3.fritz.box,unknown,Small (1k×5),v2,1000,5,0.290,0.290,89.06,
+2025-10-25T15:45:17.308927,Marians-MBP-3.fritz.box,unknown,Small (1k×5),v4,1000,5,0.001,0.001,17705.52,
+2025-10-25T15:45:17.308931,Marians-MBP-3.fritz.box,unknown,Medium (5k×5),robust,5000,5,126.362,25.272,1.00,
+2025-10-25T15:45:17.308934,Marians-MBP-3.fritz.box,unknown,Medium (5k×5),v2,5000,5,1.497,0.299,84.43,
+2025-10-25T15:45:17.308938,Marians-MBP-3.fritz.box,unknown,Medium (5k×5),v4,5000,5,0.003,0.001,41552.90,

UTILS/dfextensions/groupby_regression/benchmark_results/latest/comparison_report.txt

@@ -0,0 +1,38 @@
+GroupBy Regression: Engine Comparison
+======================================================================
+Date: 2025-10-25T15:45:17.308438
+Host: Marians-MBP-3.fritz.box
+
+
+Scenario: Tiny (100×5)
+Dataset: 100 groups × 5 rows
+----------------------------------------------------------------------
+robust  :   2.68s ( 26.80s/1k) [100 groups]
+v2      :   0.78s (  7.78s/1k) [100 groups]
+v4      :   0.68s (  6.84s/1k) [100 groups]
+
+Speedup vs robust:
+  v2: 3.4×
+  v4: 3.9×
+
+Scenario: Small (1k×5)
+Dataset: 1000 groups × 5 rows
+----------------------------------------------------------------------
+robust  :  25.85s ( 25.85s/1k) [1000 groups]
+v2      :   0.29s (  0.29s/1k) [1000 groups]
+v4      :   0.00s (  0.00s/1k) [1000 groups]
+
+Speedup vs robust:
+  v2: 89.1×
+  v4: 17705.5×
+
+Scenario: Medium (5k×5)
+Dataset: 5000 groups × 5 rows
+----------------------------------------------------------------------
+robust  : 126.36s ( 25.27s/1k) [5000 groups]
+v2      :   1.50s (  0.30s/1k) [5000 groups]
+v4      :   0.00s (  0.00s/1k) [5000 groups]
+
+Speedup vs robust:
+  v2: 84.4×
+  v4: 41552.9×

UTILS/dfextensions/groupby_regression/benchmark_results/latest/comparison_results.csv

@@ -0,0 +1,10 @@
+scenario,engine,n_groups,rows_per_group,duration_s,per_1k_s,n_groups_actual,speedup
+Tiny (100×5),robust,100,5,2.680,26.797,100,1.00
+Tiny (100×5),v2,100,5,0.778,7.777,100,3.45
+Tiny (100×5),v4,100,5,0.684,6.840,100,3.92
+Small (1k×5),robust,1000,5,25.852,25.852,1000,1.00
+Small (1k×5),v2,1000,5,0.290,0.290,1000,89.06
+Small (1k×5),v4,1000,5,0.001,0.001,1000,17705.52
+Medium (5k×5),robust,5000,5,126.362,25.272,5000,1.00
+Medium (5k×5),v2,5000,5,1.497,0.299,5000,84.43
+Medium (5k×5),v4,5000,5,0.003,0.001,5000,41552.90

UTILS/dfextensions/groupby_regression/benchmarks/bench_comparison.py

@@ -10,6 +10,12 @@
 Usage:
     python bench_comparison.py
     python bench_comparison.py --scenarios all  # More scenarios
+
+# IMPORTANT: Robust implementation is very slow on small groups (< 50 rows/group)
+# Quick mode limited to 2k groups max to keep runtime reasonable.
+# For small group sizes, use optimized implementations (v2/v3/v4).
+# Robust is designed for large groups with robust statistics needs.
+
 """
 
 import argparse
@@ -379,7 +385,7 @@ def main():
         scenarios = [
             ("Tiny (100×5)", 100, 5),
             ("Small (1k×5)", 1000, 5),
-            ("Medium (5k×5)", 5000, 5),
+            ("Medium (5k×5)", 2000, 5),
         ]
     else:  # all
         scenarios = [