Improve error warning for dist_cp loading without optimizer state

[j316chuck]

What does this PR do?

Improve error logging for models saved with load_weights_only=True and loaded with load_weights_only=False under the sharded checkpointing code path.

What issue(s) does this change relate to?

https://databricks.atlassian.net/browse/GRT-2801

Tests

Before: 1-node-mpt-13b-monolithic-crusoe-EVyT86 - optimizer key error 🔴

[rank4]: "/usr/lib/python3/dist-packages/torch/distributed/checkpoint/default_planner.py"
[rank4]: , line 354, in create_default_local_load_plan
[rank4]:     raise RuntimeError(f"Missing key in checkpoint state_dict: {fqn}.")
[rank4]: RuntimeError: Missing key in checkpoint state_dict:
[rank4]: state.model.model.transformer.blocks.5.norm_1.weight.

After: 1-node-mpt-13b-monolithic-crusoe-AF3tW4 - proper error warning ✅
then error about optimizer state is thrown again

2025-01-24 01:12:52,470: rank0[462][MainThread]: INFO: composer.utils.checkpoint: Optimizer states are not in the state_dict and won\'t be loaded. 
2025-01-24 01:12:52,470: rank0[462][MainThread]: INFO: Consider setting load_weights_only=True or ensure that the optimizer state is saved in the checkpoint.