[CLEANUP] Refactor DataAdapter to not use observers or array observers

[pzuraq]

The DataAdapter class is exposed specifically for the Ember Inspector to
support Ember Data. Currently, the adapter relies on both array
observers and standard observers. This PR refactors the adapter to use
autotracking-based methods for tracking changes instead, with minimal
API churn.

The main change is to introduce a regular poll that checks for changes
after each backburner runloop. There are two types of polling done:

1. TypeWatcher, which checks to see if a record array has changed at
   all (e.g. contents have changed)
2. RecordsWatcher, which watches the array as a whole and the
   individual records within it. Whenever any changes occur, it
   reiterates the records array, and looks for new records, updated
   records, and removed records, and notifies of any changes.

This removes the need for both array observers and standard observers.
Standard observers were previously used to detect changes on the model
instances, but now we autotrack serializing them, and that
autotracking is used to detect if they've been updated.

## Breaking Changes

• The recordsRemoved callback from watchRecords has changed. It now
  receives an array of removed records, not an index and count. On the
  consuming side in Ember Inspector, we'll have to update the code to
  remove the items passed via id explicitly.
• The observeRecord method has been removed. Because serializing the
  object is observing it (via autotracking) there is no need to
  manually observe it. Users who are still implementing this method
  (e.g. Ember Data) can still keep it, and it shouldn't cause issues, it
  just will no longer be called.

[rwjblue]

I left a few inline comments, but we also need to change things around to avoid adding more than one backburner.on('end', ...) callback as well as ensuring that we clean it up upon destruction. As it stands now, we leak in a few ways (registering the same callback many times, never removing but adding another callback per-container, etc).

[pzuraq]

This file was partially sorted, but not fully. I decided to sort it to make it consistent.

[rwjblue]

Still doesn't seem to backburner.off('end', this.flushWatchers) upon destruction (which means it leaks across multiple owners / tests).

[snewcomer]

https://github.com/emberjs/data/pull/7435/files

We have a minor test issue as a result of this refactor. Any ideas on how we can wait for the flush to happen? I see you are taking advantage of runLoopSettled in the tests here...

[pzuraq]

Correct, you should be able to use the settled() helper from @ember/test-helpers, which is essentially the same. I believe I tested this out locally before we merged this change

[snewcomer]

Oh jeez. I was putting the settled() after the async action but before set. Now it works with it in the right place. Thanks!