feat: Provide normalizeDepth option and sensible default for scope methods

[kamilogorek]

The initial issue wasn't with an incorrect circular algorithm implementation. It was with recursive references with a large payload, that were outgrowing the stack.

The most basic way to crash it:

class C {
  constructor(count) {
    if (count > 1000) return;
    this.tail = new C(count + 1);
  }
  data = Array.from({ length: 1000 }).fill('A'.repeat(1000));
}

Sentry.normalize(new C(0)));

The only notable change here is a conflict with ExtractErrorData that keeps error details one level deeper than regular "scope data", eg. contexts.errorName.data, not contexts.key so it's own depth option will be effectively ignored when normalizeDepth value is smaller.

Similar thing would happen for console breadcrumbs integration, as it keeps the data in breadcrumb.data.arguments, but because I only normalize breadcrumb.data, it doesn't affect it.

Fixes #2178
Fixes #2311
Fixes #2359
Fixes #2366
Fixes #2375

[getsentry-bot]

	Warnings
⚠️	Please add a changelog entry for your changes.

	Messages
📖	✅ TSLint passed
📖	@sentry/browser bundle gzip'ed minified size: (ES5: 16.6855 kB) (ES6: 15.6865 kB)

Generated by 🚫 dangerJS against f0a2d9a

[lobsterkatie]

👍🏻

[kamilogorek]

This is "kinda" breaking if someone is relying on such an odd data point, but if we won't change it, it'll have same issue as the one described for ExtractErrorData

[kamilogorek]

The implementation here was incorrect from the beginning. normalize returned any, so we did this._user = <any>, which shouldn't be allowed, as this._user is a non-optional User type. Also, clear sets it to {} not null.

[lobsterkatie]

LGTM!

[HazAT]

One nit + Changelog please.
Otherwise LGTM