[GH-3218] Support integer values in Moran's I

[jiayuasu]

What changed

• Normalize the selected Moran value column to DOUBLE before computing the mean and centered-value statistics.
• Build mean, centering, and neighbor expressions with Spark's column API instead of interpolated SQL, including a typed literal for the mean.
• Treat configured ID and value column names as literal top-level and nested field names, including names containing dots.
• Reject all-null value columns and non-finite means with clear IllegalArgumentException messages.
• Add regression coverage for BIGINT values, dotted column names, all-null values, and NaN means.

Why

Moran's I returns its intermediate aggregates through a typed tuple of Double values. With an integral input column, Spark inferred z * z and its sum as decimal values, then rejected the attempted upcast from DECIMAL(38, scale) to DOUBLE.

Normalizing the value column at the calculation boundary keeps the intermediate expressions consistent with the existing double-based result contract. Callers can now pass integer counts without adding their own cast.

The same calculation interpolated the mean and configured column names into SQL expressions. That made null or non-finite means fragile and caused dots in literal column names to be parsed as nested paths. Column expressions preserve the intended types and field names throughout the calculation.

Validation

• MoranTest on Spark 3.4 / Scala 2.12: 8 tests passed.
• MoranTest on Spark 3.5.8 / Scala 2.12: 8 tests passed.
• MoranTest on Spark 4.1.1 / Scala 2.13: 8 tests passed.
• Repository pre-commit hooks passed.

Fixes #3218.