Add option to use original Rubin (1987) degrees-of-freedom pooling

Feature Request

Currently, the rbmi package uses the Barnard-Rubin (1999) method for pooling degrees of freedom when combining results from multiple imputations (see rbmi:::rubin_df()). This is an appropriate default and is statistically preferable in many situations, particularly for small samples.

However, it would be useful to provide an option to use the original Rubin (1987) degrees-of-freedom method as an alternative.

The Barnard-Rubin adjustment should remain the default, with the original Rubin method available as an explicit user-selectable option.

Motivation

Some legacy analyses (e.g., SAS PROC MIANALYZE without the EDF option) use the original Rubin pooling approach. Supporting this method in rbmi would allow users to:

1. Reproduce historical analyses and validate R results against legacy SAS outputs.
2. Facilitate migration from SAS to R while maintaining consistency with previously generated study outputs.
3. Compare results obtained using the original Rubin and Barnard-Rubin methods when required.

Details