358e3a60

By: Nic Fink <n.fink@bluewin.ch>

Improve sample and campaign panels; fix checkpoint recovery stalls

- Distinguish complex norm and absolute component means, and calculate RSD
  and ESS from the weighted training projection, including legacy snapshots.
- Reorder sample results and plots, expose raw integrand and gammaloop
  Jacobian values, and add scientific formatting and full-precision points.
- Reorganize campaign panels and add sortable, numbered sub-runs with
  absolute component estimates, their uncertainties, and relative errors.
- Normalize checkpoint JSON numbers before recovery comparison so large
  Jacobians cannot trigger endless activation retries; retain conflict guards.
- Exclude expired worker leases from lifecycle counts and distinguish
  initializing samplers from active samplers in campaign sub-run status.
- Add regression coverage, operational documentation, and the campaign 55
  investigation with read-only diagnostic queries.

Validation: 296 library tests, 30 PostgreSQL integration tests, campaign
recovery after sampler and evaluator loss, and Clippy with warnings denied.