virtualization: properly restore cold/warm states, cache, and scry stack when catching an error

[eamsden]

Fixed a bug caught by @ashelkovnykov

@ashelkovnykov please confirm my understanding. In particular:

• jet_mute and jet_mure had the same bug as mink
• jet_mute and jet_mure additionally did not properly reset the scry stack in the context when catching an error.
• (not changed yet) mink further does not properly reset the scry stack in the context when catching an error

[ashelkovnykov]

• jet_mure and jet_mute were also not saving Cold/Warm state correctly
• jet_mure and jet_mute did not correctly reset the scry stack on deterministic error & blocked scry
• mink does reset the scry stack correctly in all cases (via jet_mink)

[eamsden]

Mostly just: "shouldn't we restore the saved cache after failure?" in a bunch of different places. I might not be following something...

[eamsden]

Notes from discussion with @ashelkovnykov this morning

• interpret() should restore everything except the scry stack on error (since mink needs to check the scry stack as of the error to see if it should re-throw scry-crashed or convert to another error
• All virtualization functions should call util::mink
• I need to review that leaving the scry stack in place does not cause use-after-free

[ashelkovnykov]

Follow-up to @eamsden 's comments above: I tried to elide several calls and unnecessary restorations in jet_mink, jet_mure, jet_mute, but it's making more problems than it's solving. It's resulting in the logic getting slightly more spread out, but also - much worse than that - it's making the logic of when we do/don't have a valid cold state / warm state / cache / scry stack hard to follow.

Therefore, we're going to put all of the logic (except scry stack) into interpret itself, and then force all possible virtualization through util::mink. Then we can verify all of the Context is in the correct state by checking only three locations: jets::nock::util::mink, interpreter::interpret, and interpreter::exit.

[eamsden]

This looks good to me now, but it needs to be tested booting a fakezod and someone not me or @ashelkovnykov needs to review it.

[ashelkovnykov]

@eamsden Currently failing on fakezod boot

[joemfb]

this looks consistent to me