Add exclude_unchanged parameter to dft_rast_transition() to drop same-class transitions

[NewGraphEnvironment]

Problem

dft_rast_transition() with from_class set keeps all pixels that started in that class — including ones that stayed in the same class. For example, from_class = "Trees" returns Trees→Trees (stable), Trees→Rangeland, Trees→Water, etc. all in the same raster.

For change detection workflows the unchanged transitions are noise — users want to see what actually changed. Currently they have to filter downstream:

• In dft_transition_vectors() output: transition != "Trees -> Trees"
• In QGIS: same filter on the attribute table
• In summary tables: drop rows where from_class == to_class

In a real example (restoration_wedzin_kwa_2024), 12,960 ha of Trees→Trees patches dominate the output while only ~1,100 ha of actual change is what matters.

Proposed Solution

Add an exclude_unchanged parameter to dft_rast_transition() (and dft_transition_vectors() could inherit or expose its own):

dft_rast_transition(x, from, to, from_class = NULL, to_class = NULL,
                    exclude_unchanged = FALSE, ...)

When TRUE, pixels where from_class == to_class are masked to NA before computing the summary and patches. Default FALSE preserves current behavior.

Same pattern in dft_transition_vectors() if it doesn't naturally inherit.

Why default FALSE

Backward compatible. Users opt in to filtering. Some workflows do want to see stable areas alongside change (e.g., reporting "X ha of Trees remained Trees, Y ha converted").

Alternative

Default TRUE and add include_unchanged = FALSE for users who want stable areas. Cleaner default for the common case (change detection) but breaks existing code.

Relates to NewGraphEnvironment/restoration_wedzin_kwa_2024#142