[Tree unification 3a/9] Rzepakowski normalization for KL/ED/Chi + CTS criterion

[jeongyoonlee]

Add the Rzepakowski normalization factor (normalization=True is the default) to the KL/ED/Chi criteria, plus the CTS (Contextual Treatment Selection) criterion.

Depends on Issue 1 (KL/ED/Chi scaffold). Grouped with Issue 3b (the two-class/variance criteria) under "remaining criteria".

Scope

• Rzepakowski normalization normI / arr_normI (uplift.pyx ~1625–1734) applied to KL/ED/Chi so the gain is divided by the treatment-vs-control distribution divergence. Respect the normalization flag (default True).
• CTS criterion (simple multi-treatment gain: max over groups of the outcome mean).

Verify

• Per-criterion parity vs legacy with normalization=True on fixed seeds (KL/ED/Chi).
• CTS parity vs UpliftTreeClassifier(evaluationFunction="CTS") on a fixed seed.

Reusable templates

• Legacy: uplift.pyx normI/arr_normI (~1625–1734), evaluate_CTS.

Part of #945.