Add message output estimation

[MujkicA]

Closes #845

Extends estimate_tx_dependancies to also add output messages if they're missing.

[digorithm]

LGTM -- just left a few clarification questions!

[segfault-magnet]

Reviewed alongside @hal3e. We've put our bigger impacting suggestions in a separate branch, you can compare the differences here.

Reasons for the suggestions:

1. If a revert happened not because of a missing variable or message output then the code assumes that it must have been a missing contract input. This isn't true.
2. The message inside expect_err in the abovementioned case will never be propagated. The result in question will always be an Err at that place in the code so expect_err will never panic. If the original intent was for the message to be propagated then our changes offer one way of achieving that.
3. The current code assumes there can only be one revert reason (since only one match branch can be selected). Although this is probably true atm, it costs nothing to write it in a manner that can support multiple reverts.

We haven't discussed the MultiCallHandler's estimations. Perhaps they might benefit from the same suggestions.

[iqdecay]

LGTM

[Salka1988]

👍