Make devices and presences lack property names

[michaelfig]

Closes #792

This PR improves the debuggability and predictability of device nodes and presences created by SwingSet and CapTP. Adding a Symbol.for('SwingSet') property (which doesn't show up with Object.getOwnPropertyNames), and setting a prototype that defines toString() (which allows the presence to be stringified as before) means that empty objects are marshalled compatibly as empty presences.

The REPL is improved to sniff specifically for this mechanism to distinguish presences from empty objects.

[warner]

Ugh, tap is starting to annoy me.. it's really hard to find where the test failure is.

Looks like it's in SwingSet, in test-marshal.js. I'm guessing we have a test that expects .toString to be present.

[michaelfig]

I've updated the approach: the current idea is to represent presences as an empty object with a prototype that defines:

{
  toString() { ... },
  [Symbol.for('SwingSet')]: true,
  [Symbol.toStringTag]: description,
}

This allows for simple identification of the objects in the REPL, as well as informative printing and stringification.

[michaelfig]

For example:

> const o = ...made as described;
> o
Object [Presence 52] {}
> String(o)
'[Presence 52]'
> throw o
Uncaught Object [Presence 52] {}
> Object.getOwnPropertyNames(o)
[]
> Object.getOwnPropertySymbols(o)
[]
>

[warner]

If we're going to implement #792, this is a reasonable way to do it. But please get @erights 's approval on the overall approach before landing, he's got a better sense of what sorts of promise we ought to be making than I do.

[erights]

Is this subsumed by #804 ?

[michaelfig]

Is this subsumed by #804 ?

Yes.