[BUG] Incorrect F-test denominator assignments in TwoWayAnovaModel

[LibraChris]

Describe the bug

In the current implementation of TwoWayAnovaModel, the logic for assigning denominator mean squares in F-test creation is incorrect for all three model types. This leads to incorrect results when performing two-way ANOVA.

Steps to reproduce the behavior:

Compare two way anova results to implementations of R or python

Expected Behavior

Each TwoWayAnovaModel type should use the correct denominator mean square for its F-tests:

• Fixed: denominator = error mean squares
• Random: denominator = factor mean squares (depending on nesting)
• Mixed: denominator = error mean squares for fixed factor, factor mean squares for random factor

Actual Behavior

The denominator assignments are mismatched:

• Fixed case is using Random logic
• Random case is using Mixed logic
• Mixed case is using Fixed logic

This causes incorrect F-test values and potentially invalid conclusions.