Since #234, boost-cpp is not managed by the global pinning here, which looks a bit risky. boost-cpp does not have a run-export, so thisn't viral in terms of incompatibility at least, but depending on which parts of boost are used, this could still bite us.

Unpinning boost would mean having yet more migration PRs, so if we know we're using the header-only portions, keeping the lower bound might actually be better.

Thoughts @conda-forge/arrow-cpp?