Dataset Representation Refactor

[sslivkoff]

This PR updates the way that cryo defines datasets and performs orchestration, significantly simplifying each

Highlights

1. Extraction logic is now entirely separate from transformation logic. Before, each dataset definition had to specify large, ugly, deeply-nested functions to perform extraction and transformation. Now, each dataset just defines very simple extract and transform functions, and makes no mention of tokio.
2. Dataset-agnostic processes like orchestration, partitioning, and dataframe creation are also abstracted away from each dataset definition. These processes are nearly identical for each dataset, so they are now handled by dataset-agnostic functions and macros (store! and cryo_to_df::to_df).
3. Instead of just partitioning datasets by blocks or by tx hashes, datasets can now be partitioned by any relevant dataset parameter including addresses, call_datas, or log topics. This is made possible and simple thanks to the changes in (1) and (2).

This cleans up the ugliest part of the cryo codebase. The code is now:

• significantly shorter and simpler
• easier to read, maintain, and test
• more robust
• more extensible for new features and new contributors

Code guide

• Each dataset now defines its extraction and transformation logic using the CollectByBlock and/or CollectByTransaction traits.
• Each dataset simply defines these parameters.
• A new crate cryo_to_df defines a procedural macro that adds a to_df() function to each ColumnData.