Reliably generate mutations while the snapshot is iterating

[eoghanmurray]

Reliably generate mutations while the snapshot is iterating in order to test rarely occurring problems. Have added what I believe should be the correct parsing, on the assumption that the DOM changes will subsequently be correctly picked up by mutation events.

• The difference between test cases a & b demonstrate different outcomes because of the freezing of childNodes in for (const childN of Array.from(n.childNodes)) — I'm not sure if there's a better way to iterate over child nodes (I imagine this freezing was intentional)
• Test case d takes this further to show that it should not be possible to iterate to the #d3 element
• The final e test case is the important one as it demonstrates that the snapshot can produce duplicate nodes, in this case #a1 appears twice in the output

I authored #60 assuming there was something similar to the e test case going on.
But maybe a better solution would be to ensure all __sn attributes are wiped before the start of a snapshot, and ignore any nodes that already have a __sn assigned if we encounter them in the DOM walk, as presumably they have a lower id from earlier in the walk.

Node: The code modifications in snapshot.ts to get these tests to work would need to be removed or somehow only targetted to a special test build.

I'd like to get further input into this before proceeding on fixes for the test case failures I've created in this pull request!

[eoghanmurray]

@Yuyz0112 Just to note that it is intentional that the tests are failing for this pull request (in case you are avoiding it for that reason).

I also need some good ideas as to how to target the bit of code which does the eval in d6b57fc#diff-75923536d00ea7b9bdbd2a5196a698720893adb4993f502c3d777ca9b51a4df2 so that it only executes in testing?