fix: properly restore nested wildcard values

[matt-clarson]

fixes #50
also fixes pinojs/pino#1745

This fixes the issue of restoring redacting values across multiple nested wildcards - the problem was that the redact/restore modifiers assumed the same value could be used to restore all instances of a redacted path. To fix, I've changed the specialSet function to keep track of the paths it visits in a tree, so that the correct path/value can be stored for use in restoring.

This has required a couple of extra modifications to the specialSet and iterateNthLevel functions:

• special set now doesn't return anything - it takes the store as an argument and pushes restore instructions to it directly
  • this also means that there's no need to return an exists value for paths that aren't being redacted - where that was being done we can just not update the store
• the iterateNthLevel function also doesn't returning anything now, similarly taking and mutating the store itself
  • as far as i can tell, this change also fixed a separate bug where iterateNthLevel would exit early, leaving some parts of a wildcard path untouched - this is covered in a new test case

[simoneb]

Very hard to make sense of what's going on here without intimate knowledge of this package, so I'll just comment for now.

I'm fairly confident that this package has extensive code coverage and therefore, if all tests pass in addition to the new ones, you fixed the issue without introducing any regression, but the package maintainer is probably the best person to confirm that.

Apart from this, excellent job, the code is quite cryptic and solving the issue without introducing regressions is a remarkable achievement.

[mcollina]

lgtm

[mcollina]

this is amazing work, thanks!