POET observations - BREAK INTO ISSUES

[avive]

1. Prover should only use o(n) labels in ram while computing dag leaves - no need to read from the dag storage file
2. There should be only one binary storage file for the labels so it is easy to build a proof for large values of n by looking up label values in the storage file
3. Prover should use buffered file i/o with a large ram buffer to avoid i/o from slowing down dag generation. Ideally benchmark vs i/o in another thread for large values of n
4. Proofs should not include duplicated data (label values)
5. Prover must consistently work large values of n and we should do more testing with large values of n