Keep a full CI run under one billed minute #16

A full PR run of ci.yml takes about 1m03s, just over the 60 s billing line, so it bills 2 minutes instead of 1. One test sets the floor of the parallel run: test_every_theme_paints_every_screen (~22 s).

Split it (e.g. parametrize per theme or per screen) so pytest -n auto can spread it across workers, and check that a full CI run comes in under 60 s.

Small and self-contained, a good first contribution. mise run test and mise run ci:local run the suite locally.