Make some type parameters inside ConcurrentHashMap non-null.

[cpovirk]

We already had the parameters on the class itself correct. However, we
had declarations inside incorrect. This PR fixes most but not all.

For the most part, this PR doesn't matter: Most of these APIs are not
user-visible. We had them annotated in the first place only because our
CF-stub-conversion tool changes <T> to <T extends @Nullable Object>
everywhere inside @NullMarked code, without regard to whether the API
is publicly visible. In fairness, this makes some sense because a
non-publicly visible type can be extended by a publicly visible type,
which can render the non-publicly visible type's APIs suddenly publicly
visible. In fact, that happens in this very file: The package-private
CollectionView is extended by the public KeySetView. Still, for
most of the non-visible APIs, this PR has no actual effect: It's just
a simplification (by removing information about non-visible APIs, which
we don't normally annotate) that happens to also make those APIs'
annotations more correct.

For public APIs like KeySetView itself, it would have made sense for
the CF stubs to have <T extends Object>, which would have translated
into plain <T> in our stubs. The PR's change to <T> could affect
existing callers.

I will be back with another PR that handles some other type parameters
as part of actually annotating some other unannotated APIs in the class.

[wmdietl]

Null-hostility looks good, just one place would benefit from some comment.

[wmdietl]

The comment above says Subclasses of Node with a negative hash field are special, and contain null keys and values.
To account for these subclasses, the type parameters here should allow null.
If you did consider this, could you add a CFComment why this is better?

[cpovirk]

Thanks, I hadn't. This change was just part of my mission to remove annotations from APIs that aren't visible to users. I left some new notes about that in #101. Let me know whether you have thoughts on that approach, whether this type is exposed through a subtype you missed, or if there's anything else I might be overlooking here.