NumCosmoHIPERT_BOLTZMANN_REQUIRE_LMIN

Smallest multipole nc_hipert_boltzmann_require() will ask the Boltzmann code for. The lensing convolution mixes multipoles and the transfer sampling coarsens when the requested range is small, so a Boltzmann code needs headroom above the last $\ell$ used. NcCBE already adds its own margin (CLASS delta_l_max), which alone does not converge a low-$\ell$-only request: a standalone low-$\ell$ likelihood asking for $\ell_{max} = 29$ lands $\sim 0.01$ away in $-2\ln L$ from the value reached once a few hundred multipoles are requested.