fix unnecessary memory retention paths

[mhofman]

While investigating Agoric/agoric-sdk#5447, I found that endo was possibly unnecessarily holding onto memory in a couple places. This PR fixes them:

• SES's v8 Error taming no longer holds onto the raw CallSite stack for error objects after it has created a string for them. It does so by keeping a second WeakMap with the generates stack string, and nulls out the content of the first WeakMap. That way the stack is still processed lazily.
• stream-node possibly held onto the finalResolution promise and other things in context even after the stream closed.

[kriskowal]

New rule added to head canon: Law of the Conservation of Event Handles: every on must be balanced by an equal but opposite off.

[mhofman]

New rule added to head canon: Law of the Conservation of Event Handles: every on must be balanced by an equal but opposite off.

It's preferable.

@erights I'll wait for you to review the v8 Error taming changes

[erights]

On the writer.js changes, I have never used .on or .off but see they are inherited from EventEmitter, which is fine since this whole file is Node specific. But is the problem Node specific? Should we have some Writer-level cleanup that applies to all Writers, as opposed to just Writables?

[mhofman]

@erights the cleanup for iterator based streams is to use return or throw on the iterator. The thing is node can be a little finicky, and for some reason I can't recall, we had to add listeners directly on the node writable stream as well. However those have cleanup semantics which were not respected here. I'm not saying this listener cleanup was entirely necessary, just that it's always better to do an explicit cleanup if you can.

In this PR, I think the more meaningful change is to the v8 error stack taming.

[mhofman]

@erights PTAL, I decided to redo the logic to make clear we either have a raw structured stack trace or a parsed string.

[erights]

New unified map is better. LGTM