To align this with OTEL types.

This removes the public field to get/set `status` on a span/transaction.

Now we will no longer be supporting Node v8, these imports are no longer
required

This is an attempt to restructure our migration docs in preparation for
the first alpha.

When Sentry started supporting the idea of exception groups, two changes
happened. In the SDK, we adapted our logic for handling linked errors to
also handle `AggregateError`s. And in our ingest pipeline, we began
looking for an `is_exception_group` flag on the last entry in
`event.exception.values; when we found it, we'd then ignore that entry
when grouping and titling events, under the assumption that it was just
a container and therefore wasn't meaningful.

When it came to instances of `AggregateError`, this worked great. For
linked errors, however, this caused us to focus on the `cause` error
rather than the error which was actually caught, with the result that it
both threw off grouping and made for some very unhelpful titling of
issues. (See the screenshot below, in which the first three errors are,
respectively, an `UndefinedResponseBodyError`, a `RequestError`, and an
`InternalServerError`, though you'd be hard pressed to figure that out
without opening them up.)

This fixes those problems by restricting the use of the
`is_exception_group` flag to `AggregateError`s.

Note: In order to update the tests to work with this change, I had add
in consideration of the error `name` property and the corresponding
event `type` property, to match what we do in real life. To keep things
readable, there's a new mock `AggregateError` class, which I adapted all
the tests to use.

Instead, in places we need it we cast to a `SentrySpan` which still has
the things in place, for now.

With this, our span interface is _almost_ the same as for otel spans -
missing are only:
* Aligning `traceFlags` - this has apparently been updated in OTEL to
have type `number` only, which makes this a bit easier.
* Aligning `setStatus` which has a different signature in OTEL.

I'll do these in follow ups!

The biggest work here was fixing tests - I tried to rewrite tests to do
less mocking where possible, which IMHO should cover actual
functionality better than before (e.g. in svelte).

`errorHandler` for `CanvasManager` was added in the latest rrweb, but
was not configured in our integration.

Closes #6230

It seems this is all that is left of the deprecated tracing options!

Flush replay when the feedback form is first opened instead of at submit
time

We are making this change because we have noticed a lot of feedback
replays only consist of the user submitting the feedback and not what
they did prior to submitting feedback. This may result in false
positives if users open but do not submit feedback, but this should make
replays from feedback more useful.

We want to filter out `ResizeObserver` error since they are not
actionable and have no stacktrace and break transaction status. ref
https://stackoverflow.com/a/77680580/1139707

Also, I added to filter out undefined errors - here are two events to
Sentry to represent this.

undefined: 

https://sentry.sentry.io/issues/3611187513/events/46ed8c398c234ff89baee87c5c341844/

ResizeObserver: 

https://sentry.sentry.io/issues/3611187513/events/48f25ea9dfbf4bd0b84a18982ee73362/

@JonasBa when you got some time, could you look into actually rewriting
this to the new style? Right now a lot of the tests depend on the
non-hook based version etc. and I'm not 100% clear on what can/cannot be
removed there 😅

The end result should be just having a single `nodeProfilingIntegration`
that receives the client in `setup(client)` and sets up hooks etc in
there!

We forgot to do that, apparently :grimace:

Co-authored-by: Lukas Stracke <lukas.stracke@sentry.io>