brain_deer.commands.warpdrive

WarpDrive commands: apply a manual-refinement session to a normalization.

These are headless and file-based, so Lead-Lizard (or any host) can drive them through --commands-json without a GUI: point at a session directory holding source.json/target.json and the two ANTs warps, and get corrected warps written to an output directory.

Replay-first by design (see the WarpDrive plan): the point pairs are the authoritative, re-appliable state, and _apply() writes new files rather than overwriting in place. Overwriting the subject’s warps is a separate, explicitly destructive warpdrive_harden step.