Picks one of the bundled Visium-capture-to-Visium-capture pairs (from Kairns' cached
experiment data), prescales the source capture onto the target's coordinate frame, then runs the real
vendored LDDMM registration — same code path as bin/align_points.py on the CLI, launched
as a background job so status can be polled across gunicorn's worker processes.
bin/align_points.py's own — 150 LDDMM iterations, the
first 75 affine-only before the non-rigid warp phase starts.bin/align_points.py's pipeline — the static
text explains what the stage does and how to read its output in general; the monospace lines under it
(once reached) are this run's actual numbers.