Re-enable the 19-contact Coax testset (currently fully commented out)

(Draft from the 2026-07 code review; posted on behalf of the review.)

The true-coaxial example with 19 contacts has its entire testset commented out
in `test/test_real_detectors.jl`, presumably because it was too slow at the
original resolution. As a result the many-contact code paths (weighting
potentials for a large contact count, per-contact grouping, the associated
plotting-free bookkeeping) are exercised nowhere in the suite — the review found
this is the only example exercising that regime.

Proposal: re-enable it on a deliberately coarse grid with loose tolerances
(e.g. only a coarse `max_tick_distance`, no refinement, asserting just
convergence, depletion status and total signal-charge sanity rather than tight
physics numbers). The point is that the 19-contact path runs at all, not that it
is precise. If even the coarse version is too slow for CI, running it only on
one CI job (nightly / one OS) would still be better than nothing.

Start with the fully commented 19-contact testset in test/test_real_detectors.jl and compare its current setup with the proposed coarse-grid settings. Run the test with loose tolerances and verify that it checks convergence, depletion status, and total signal-charge sanity; done means the many-contact path executes reliably in an appropriate CI job.

Written by the indexing model from the issue text.