use Chainwatch and Sentinel to identify worthy real-life test vectors

[raulk]

Analyse chain data from real networks (test net, calibration net, incentivised testnet, etc.) in order to to identify and cherry-pick messages, blocks, tipsets, that trigger unique conditions / transformations / state transitions / execution paths that we want to lock into a test, and eventually turn into a regression test for future replay.

This enables us to seed our Test Vector Corpus with useful entries from the get-go.

Example conditions to target:

• Happy path for all system actors.
• Messages that led to unique errors for all system actors (dig into receipts).
• Out-of-gas messages.
• Syntactically invalid messages.
• Messages with/without parameters.
• Tipsets that include duplicate messages.
• Tipsets that include semantically equivalent blocks.
• etc.