Add non-baseline example submissions

We should demonstrate the capabilities that our API gives users, to give them examples of what they can explore when developing submissions. We should make some example submissions, probably in a new reference_algorithms/examples dir that we link to from README.md. These could include:

• gradient accumulation
• data selection
• different LRs on different layer types (embeddings and norm layers for example)

or whatever other creative ideas you can come up with. The examples can be simple forks of the NAdamW baseline.

Start by reading README.md and the existing NAdamW baseline in reference_algorithms to understand how submissions are structured. Add several simple example submissions, potentially covering gradient accumulation, data selection, or different learning rates for embeddings and normalization layers, in a new reference_algorithms/examples directory, and link that directory from README.md.