Include missing values as a separate PSI bin

[lcrmorin]

Automatic Handling missing values in PSI is important for me.

I offer the PR below - mostly AI generated and human reviewed.

---

Closes #883. Adds opt-in missing_values='include' to measure missingness drift
without merging missing observations into real-value bins. Numeric boundaries are
learned from observed basis values; all rows enter the frequency denominator.
Existing raise and ignore behaviour is preserved.

Proposed edge-case policy: with fewer than two distinct observed basis values,
compare a missing bin with a single observed bin rather than learn boundaries
from the test partition. Automatic thresholds count the populated bins, including
missingness. This policy remains open to maintainer feedback.

Validation: 77 focused tests pass on pandas 3.0/sklearn 1.9 and pandas 2.2/sklearn 1.4.
Tests include hand-calculated PSI, all-missing partitions, automatic thresholds,
duplicate row labels and category-name collisions. User guide and API docs updated;
type and style checks pass.

Dependency requirements are unchanged. Testing covers the two configurations above, not the full supported-version matrix.