Implemented rendering test results for the runner.

There are two report types that are specifiable in the program args:

• Generate a summary markdown report for each test group with a row for the number of tests that passed for each sub-group.
• Run any tests that match a given (optional) string filter and output all tests that match. groups and sub-groups are flattened and all the individual tests that match are kept. Reports if the test passed and displays any failure information (if any). The output is rendered as markdown to stdout.

I think this is one of those cases where writing unit tests is not really worth it (just because we would have to write tests that match against generated markdown, which is probably going to be fairly unstable). So we'll work out the bugs once we are reading in data from the parser.

Also updated IncorrectAccountFinalState to show which trie hashes did not match in the case that at least one hash differs.

This PR removes the imperative deserialization logic in favor of the declarative and strongly typed struct style enabled by serde.

You'll note the

#![allow(dead_code)]

Attribute in the deserialization module. We can remove the unused fields once the trie generation is implemented; disabling the warnings for now.

Implements reading and running tests against Plonky2. The last bit of generating the report still needs to be done, but I suspect that it will increase the PR size a bit, so I make that a separate PR.

This PR removes all the logic associated with incremental parsing via commit date. Until / unless parsing proves to be a really slow process, it is favorable to keep the logic simple. In other words, we re-parse everything on invocation of the parsing binary until the performance becomes an issue.

Additionally, this PR modifies the cloning logic to leverage sparse-checkout functionality such that it only fetches the folders and test cases that we care about. This dramatically reduces the clone time, to roughly 7 seconds from 45 seconds on a gigabit connection.