Update Actors error handling

[austinabell]

Summary of changes
Changes introduced in this pull request:

• This all stems from the gas charging blockstore (and some from gas charging syscalls). In the go implementation, they panic exactly where the error is encountered and that error is caught outside of their vm. Since we are (and should) bubble the errors up instead of panicking, that error exit code cannot be lost. This is done through the error downcasting. The reason for the trait implementation is because Amt and Hamt errors, as well as dynamic errors can downcast into a ActorError because of SysErrOutOfGas from blockstore interactions within those structures.
  • I've checked every mapped error and conversion (there is over 500) and this is what I've come up with

The only time this can ever be a difference is if the go implementation uses (ExitCode) Wrapf or rt.Abort, an out of gas error was not hit, and there was a different actor error was hit that would be overwritten by this. After checking all usages of this (I could be missing a case and was just skimming, there was ~300 of these as well) I'm semi confident all of them do not fit this criteria, or when they do the go implementation would just fatal error (so it doesn't matter)

Reference issue to close (if applicable)

Closes #720

Other information and links

[timvermeulen]

Good stuff! The ActorDowncast trait looks like a good approach to me.

[dutterbutter]

Optional/Nit: 'replace' instead of replaces