Define __slots__ for small classes

[lgeiger]

TensorFlow has many small classes that are used internally as wrappers or containers. Frequently used objects can benefit from defining __slots__.

From the Python docs:

__slots__ allow us to explicitly declare data members (like properties) and deny the creation of __dict__ and __weakref__ (unless explicitly declared in __slots__ or available in a parent.)
The space saved over using __dict__ can be significant. Attribute lookup speed can be significantly improved as well.

This PR explicitely defines __slots__ for a non exhaustive list of classes that have very few attributes, tend not to be inherited from and don't use dynamic assignment of new variables.

[lgeiger]

Sorry about the failing tests (it's a bit challenging to run all test locally without a proper bazel cache) af273bc should fix it.

[lgeiger]

aa41fec should fix the last CI breakage, sorry about that.