Fix signature verification in mempool

[austinabell]

Summary of changes
Changes introduced in this pull request:

• Fixes signature verification in new_from_parts
• Fixes signature generation and verification in message pool
• Released new versions
  • Only message crate was a breaking change, and removed the types dependency from the crate (to avoid fil-proofs from being pulled into this dep tree)
  • Added CHANGELOG to message crate, to allow @rllola and @jleni to easily follow changes in releases for their usages.
• Cleaned up and added convenience functions for verifications to make this harder to be an issue later
  • Didn't change the new_from_parts function to not verify, because it's a good sanity check for a bit, but it probably should be removed later as there is no need for it to verify.

The issue was that message Cbor was being signed, not the message cid bytes. The only borked usage was in the mempool, which is why we didn't run into this yet. I could have sworn I commented to change this, but I guess I didn't follow up.

These interfaces can definitely be cleaned up (especially in the mempool), but I'm just going to keep scope small for this PR

Reference issue to close (if applicable)

Closes #725

Other information and links

[dutterbutter]

Should this error type be InvalidFromAddr rather than Other? It seems the verify fn would fail if the from address is not the correct address which seems more of an InvalidFromAddr error type.

[austinabell]

That's not the only error that can happen with signature validation. You also lose the error message for no benefit.

Side note, I don't like the error type in this crate, would be nice to improve later.

[ec2]

I don't get why you moved valid_for_block_inclusion from being a method on the Message trait into the msgpool

[austinabell]

I don't get why you moved valid_for_block_inclusion from being a method on the Message trait into the msgpool

The reason is because filecoin proofs is brought into the dependency tree of messages and it makes it unusable by others using published crate for no benefit to us (because proofs dependency only works on certain targets).

Idea is just to keep the message crate lean, and separating proofs logic by a feature was not something I was eager to do now, since there was no need.