Out of memory issue in CAM at ne1024 resolution

### What is the feature/what would you like to discuss?

CAM initialization consumes excessive memory (and time) at very high resolution (e.g. ne1024), to the point of risking out-of-memory during startup. Currently we have to undersubscribe a compute node on Derecho, which is probably unavoidable, but we are seeking ways to reduce the memory usage in CAM initialization.

A large fraction of the per-rank init footprint is the read-only lookup tables that are bit-identical on every rank — photolysis cross-sections and radiative source functions, aerosol optics tables, soil erodibility, and the global lat-lon source fields staged by `tracer_data`. Every rank allocates and holds its own copy, so the node-level cost scales with ranks-per-node while the useful information does not.

### Is there anyone in particular you want to be part of this conversation?

@jtruesdal

### Will this change (regression test) answers?

No

### Will you be implementing this enhancement yourself?

Yes