fix: Run attestation requests within a tokio runtime

[netrome]

Closes #1224

This PR solves the problem of the attestation generating logic happening outside of a tokio runtime by creating the root runtime first thing in StartCmd::run and calling .enter() to enter the runtime context and keep it for the remainder of the run() function.

Forgive me, team, for I have sinned. I have opened this PR without adding test coverage of the change. While the change is trivial and should be easy to review, I did not find any easy way to test this. And since this bug currently breaks the node, it would be nice to get the fix through before refactoring the code to make this testable.

I have created a follow-up (#1227) to ensure this part of the code is easier to test, but I fear that will be a significant effort to refactor and we might need to push it until after the soft launch. Unless we're hitting many more bugs during E2E testing at which point we can reconsider prioritizing this refactor.

Also this bug is a deja-vu. We fixed a similar issue in #612 but for some reason, the runtime guard introduced there is not around anymore. This also shows why it's important to have tests to guard against regressions. @DSharifi - was there any reason to kill the runtime guard that you recall?

[pbeza]

LGTM, but yeah — it’d be nice to find a proper way to test it so we don’t run into the same kind of bug for the third time.

[netrome]

LGTM, but yeah — it’d be nice to find a proper way to test it so we don’t run into the same kind of bug for the third time.

Amen to that. As much as I'd love to do #1227, if you have any ideas of lower-hanging fruits or quick solutions in the meantime I'd love to hear them.