fix: better tracking of seen objects in error serialization

[sam-super]

Description of the Change

Breaks circular references in before objects are serialized to prevent cryptic errors in parallel mode.
E.g. (taken from the linked issue):

 1) Uncaught error outside test suite

  0 passing (308ms)
  1 failing

  1) Uncaught error outside test suite:
     Uncaught TypeError: Converting circular structure to JSON
    --> starting at object with constructor 'Object'
    |     property 'props' -> object with constructor 'Array'
    --- index 0 closes the circle
      at stringify (<anonymous>)
      at writeChannelMessage (node:internal/child_process/serialization:120:20)
      at process.target._send (node:internal/child_process:819:17)
      at process.target.send (node:internal/child_process:719:19)
      at processTicksAndRejections (node:internal/process/task_queues:93:5)

For context we have seen this sort of error when using nestjs (e.g. when a module is missing a dependency, the error thrown contains a circular reference). It's likely something that would be good to fix in nestjs, but mocha would ideally handle it gracefully, regardless.

Alternate Designs

We could just strip all 'extra' props of the Error object and keep only the standard parts (e.g. message,name and stack). Would need to make sure those props are what we expectthem to be. This would be more performant but at the cost of losing useful information on the error object.

Why should this be in core?

Saves people from having to turn parallel mode off to work out why their test failed.

Benefits

Errors that contain circular references can be more easily debugged.

Possible Drawbacks

There is probably a performance penalty when inspecting the objects. The assumption is that it's not prohibitive.

Applicable issues

This will be a fix/patch release

Fixes #4552

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: sam-super / name: Sam Adams (171f77e, e3b3413, 4013a93, fc8076b, 695b17c, 9a7d109, c3a1129)

[coveralls]

coverage: 94.424% (+0.07%) from 94.358%
when pulling c3a1129 on sam-super:issues/4552
into 99601da on mochajs:master.

[sam-super]

Worth a re-run of the failing test?

[sam-super]

@Uzlopak looks like it might be a false positive. Windows tests ok for v14/v18 and i can't think of anything that might be low-level enough to affect just that set of versions.

[JoshuaKGoldberg]

LGTM, nice and clean! ✨

@voxpelli I know you were looking at errors too, so I'd want to defer to your review on this one too.