Unserialized errors are remote

[erights]

If linked with endojs/endo#649 , then marshal will cause unserialized errors to be identified as such when shown on the console. This tagging is invisible to unprivileged code.

If linked before that PR, then this PR has no effect. Thus, it can be merged before a SES containing that other PR is released.

Well, almost no effect. As a place holder for @dtribble's better pseudo-pseudo-random-number-generator, the remote error tagging simply starts the counter at 10000 instead of 0. This will help a bit but will still be confusing.

[erights]

I did #2773 to test whether the job stuck above in CI has anything to do with this PR.
An hour later, its
https://github.com/Agoric/agoric-sdk/pull/2773/checks?check_run_id=2240580543
is still stuck in CI, from which I conclude that whatever the problem is, it has nothing to do with this PR.

[erights]

We wanted to have the errorIdNum values generated by this PR be usefully unique, possibly paired with the errorTagNum values from endojs/endo#649 also being usefully unique. Given the Agoric sparse-int package, this all seemed trivial, until we hit tensions between various engineering principles. We agree to omit such enhanced uniqueness from these PRs, and instead file a bug on what still needs to be done: #2780

[warner]

Code changes seem fine. Consider the errorCount=0 thing but not too long.

Overall, as we figure out #2780, I'm expecting to see this converge into a form where the original error is tagged with a composite identifier that holds 1: a weakly-globally-unique-ish randomly-generated Swingset instance identifier, 2: a unique-within-that-Swingset vat ID, 3: a unique-within-that-vat Error counter. When this serialized error migrates to a different system, the Error object created during deserialization should retain all that, plus get marked with the triple of identifiers from the point of deserialization. If it travels further, I'm on the fence as to whether we accumulate the full trail of breadcrumbs, or we throw out all but the first (origin) and last.

[erights]

This PR does not provide enhanced console output until it is using a ses that incorporates endojs/endo#649 , which is not yet released, and will be after the ses 0.12.6 release. Until then, this PR doesn't break anything but also doesn't contribute much value.