Add handle ownership again

[nirinchev]

This is a refactoring of our native handle hierarchy. A summary of the changes:

• Introduces a method to release the native pointer to HandleBase<T>
  • The unowned version of the handle will not attach a native finalizer and the release method is a no-op.
• Adds RootedHandleBase<T> that is owned by a Realm and it's lifetime is at most as long as the lifetime of the parent Realm.
• Modifies all the realm types (list, results, object, query) to inherit from RootedHandleBase and modifies their ctors to optionally pass in their parent Realm as root in case it's unowned.
  • The reason we only care about the unowned case is because that's a situation where the user doesn't have explicit control over the lifetime of the Realm instance.
• RealmHandle cleans up all its children when released.

Supersedes #528. I'd like to add more meaningful tests but those need to wait until the migrations PR is in.

Fixes #527
Fixes #837

[coveralls]

Pull Request Test Coverage Report for Build 3185032259

• 136 of 144 (94.44%) changed or added relevant lines in 8 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.7%) to 87.406%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
lib/src/subscription.dart	3	4	75.0%
lib/src/scheduler.dart	0	2	0.0%
lib/src/native/realm_core.dart	98	103	95.15%

Totals
Change from base Build 3181931773:	-0.7%
Covered Lines:	1971
Relevant Lines:	2255

---

💛 - Coveralls

[nielsenko]

I think we can get rid of the ints by using an Expando. Might be worth exploring. But otherwise LGTM.

[desistefanova]

Maybe have to check also people.isManaged and people.isValid. They should't throw, right?

[desistefanova]

Do you think we may say something in the changelog about this change?

[desistefanova]

There are many places from the existing code that require two arguments, but only one is passed. Is it only on my local machine or it is because of the merge?
Please make a search for the following constructors:
RealmResultsHandle._
RealmObjectHandle._
RealmListHandle._

[nirinchev]

Do you think we may say something in the changelog about this change?

This should be purely internal, but I can add something to the internal section.

There are many places from the existing code that require two arguments, but only one is passed.

Yeah, that happened when I merged from master via the Github UI. Will fix it before merging.

[desistefanova]

Maybe this PR is the correct place to remove all the keepAlive methods. I think we don't need them anymore. @nielsenko ?

[nirinchev]

The keepAlive is necessary to make sure we don't get parts of the objects finalized mid-scope. I don't think this PR does anything that would allow us to remove them.