"On the fly" decoding

Storing all decoded logs and traces would be heavily intensive.

We could/should instead build functions that decode "on the fly" -> intake raw data and output a decoded dataframe.

A prereq is to store as many function calls, event selectors, ABIs as possible & then have some way to intelligently decide how to decode a given trace and event.

No files, tests, or entry points are named. Start by defining how function calls, event selectors, and ABIs should be stored and how a decoder should choose among them for raw traces and events. Done should mean raw input can produce a decoded dataframe without storing all decoded logs and traces.

Written by the indexing model from the issue text.