Closes #5232: _reduce improvements for ArkoudaExtensionArray

[ajpotts]

Expand ArkoudaArray._reduce to support full reduction set

Summary

This PR significantly expands the internal ArkoudaArray._reduce implementation to support a broad, pandas-compatible set of reduction operations, and adds comprehensive tests covering both functionality and argument handling.

The new implementation replaces a hard-coded conditional chain with a centralized, extensible dispatch table and introduces robust argument passthrough logic that safely forwards only supported keyword arguments to backend methods.

Key Changes

ArkoudaArray._reduce

• Accepts reduction names as either str or GroupByReductionType
• Centralizes reduction dispatch via a mapping table
• Delegates to backend methods when available, with clear error reporting otherwise
• Filters keyword arguments (skipna, ddof, etc.) based on backend method signatures
• Adds full type annotations and NumPy-style docstring sections (Parameters, Returns, Raises)

Supported reductions include:

• Numeric / boolean: all, any, sum, prod, min, max, mean, var, std, count, nunique, first
• Index-returning: argmin, argmax
• Planned / backend-dependent: median, mode, unique, or, and, xor

Reductions not implemented by the underlying Arkouda backend raise a clear TypeError.

Test Coverage

The test suite for ArkoudaArray reductions has been substantially expanded:

• Scalar reductions: verifies correct return types for all supported numeric/boolean reductions
• Argument reductions: validates argmin / argmax semantics
• Array-returning reductions: covers unique and mode
• Boolean reductions: covers or, and, xor for boolean arrays
• Keyword argument handling:
  • skipna passthrough behavior
  • ddof passthrough for var / std when supported
  • Graceful ignoring of unsupported keyword arguments

Tests automatically xfail when a reduction is not implemented by the current Arkouda backend, ensuring CI stability while preserving visibility into planned functionality.

Motivation

This change moves _reduce closer to pandas ExtensionArray expectations, reduces duplication, and provides a clear framework for incrementally adding backend support for additional reductions.

It also improves developer ergonomics by making reduction support explicit, discoverable, and well-tested.

Closes #5232: _reduce improvements for ArkoudaExtensionArray