feat: kernel: don't auto-convert ExecutionError to anyhow

[arajasek]

#1488

This PR demonstrates all of the current places where we auto-convert ExecutionErrors into fatal anyhow errors. Having looked at them all, I think we're comfortable with all of these conversions.

Next steps:

• Reviewers confirm the existing conversion sites are fine to treat as fatal
• Decide on whether we want to land this in FVM v3.
• Some way to make this more "explicitly" FATAL in code?
• Will fixup tests when we have a clearer direction.

[Stebalien]

Looking at the executor, any errors returned directly from execute_message are "super fatal" (will fail the entire block), which is different from most fatal errors (will fail a single message).

[arajasek]

Closing as stale.