Clean up GLC module: drop solve(), slim results, honest mass balance

[milanofthe]

Follow-up cleanup to the GLC/BVP1D refactor (#28).

• Remove the orphaned standalone solve() helper: it duplicated the block's BVP1D-seeding pipeline and nothing called it anymore.
• Slim the results dict from ~33 keys to 11 by dropping the blind update(phys_props)/update(dim_params) merge and unused partial-pressure keys.
• Replace the mass-balance fudge (which redistributed the residual back into the outputs to force closure) with an honest mass_balance_residual [mol/s] diagnostic. C-C closes to numerical noise (~1e-15); O-C shows the ~0.4% imbalance that the open inlet boundary genuinely leaves, instead of hiding it.

Block IO unchanged; C-C output ports bit-identical to before. Tests updated accordingly (tritium suite green).