Use CCL's CAMB backend so blinding and inference share one Boltzmann path #280

From @sachaguer's review of #241: CCL can use CAMB as its Boltzmann backend (transfer_function="boltzmann_camb" and matching halofit config). If CAMB is what the inference step runs, the blinding theory callables (#252) should call CAMB through CCL rather than using CCL's default — then blinding and inference share one power-spectrum path and the CAMB↔CCL cross-check in #252 becomes a consistency guard rather than a seam we live with.

Done when: the theory backend config in #252 exposes the Boltzmann-backend choice, the shipped default matches whatever the inference pipeline (#220) uses, and the cross-check test runs against that configuration.

— Fable on behalf of Cail.