Closes #5294: alignment tests for arkouda.numpy.random.legacy

[ajpotts]

Add NumPy-alignment tests for legacy random API (bool randint discrepancy tracked)

Summary

This PR adds a NumPy-alignment test suite covering Arkouda’s legacy random API, with an emphasis on:

• shape semantics
• value bounds
• dtype behavior
• error-message alignment with NumPy’s legacy RandomState

A known discrepancy in randint(dtype=bool) error messages is intentionally captured and marked as xfail, so the suite documents the divergence without breaking CI.

What’s included

• New test module:
  • tests/numpy/alignment_verification/random_legacy_alignment.py
• Added to pytest.ini test discovery
• Coverage for:
  • ak.random.rand (scalar and multi-dimensional shapes)
  • ak.random.randint (int64 bounds and shape semantics)
  • ak.random.uniform
  • ak.random.standard_normal (distribution sanity checks)
  • ak.random.random

NumPy alignment strategy

For randint(dtype=bool), expected error messages are derived dynamically from:

numpy.random.RandomState.randint(..., dtype=bool)

This avoids hard-coding fragile strings and ensures the tests track NumPy’s actual behavior.

Known discrepancy (xfail)

For randint(dtype=bool) with negative low:

• NumPy raises:

  ValueError: low is out of bounds for bool
  

• Arkouda currently raises:

  ValueError: low < 0
  

This mismatch is explicitly detected and marked with:

pytest.xfail(... Issue #5295)

so CI remains green while preserving a clear signal that behavior differs from NumPy.

Why this matters

• Establishes a reusable pattern for NumPy-alignment testing
• Makes random API differences explicit and actionable
• Prevents silent divergence in user-facing behavior

Follow-up

Once Arkouda’s randint(dtype=bool) validation is updated to match NumPy’s error message, the xfail can be removed and the test will pass automatically.

Closes 5294: alignment tests for arkouda.numpy.random.legacy