[FLINK-11485] [core] Refactor multiple nested serializer compatibility resolution logic to CompositeTypeSerializerUtil

This commit refactors the logic of resolving overall compatibility
results across multiple nested serializers of a composite serializer out
of the CompositeTypeSerializerSnapshot class.

This allows us to reuse this functionality when implementing the
compatibility check for the PojoSerializerSnapshot

Having "Linked" in the name better reflects the fact that the map
preserves entry ordering.

This utility class may commonly be used by the KryoSerializer and the
PojoSerializer. This commit moves the utility to a more suitable package
and adjusts usage access.

This closes #7496.

This commit adds two more tests to TypeSerializerSnapshotMigrationTestBase
that validates that the new serializer are able to read previously stored
data (perhaps after reconfiguration).

This commit adds a new KryoSerializerSnapshot class and uses that
instead of the now deprecated KryoSerializerConfigSnapshot.

It also adds migration tests for the new KryoSerializerSnapshot,
extending the TypeSerializerSnapshotMigrationTestBase.

[FLINK-11323] [tests] Adjust serializer tests to respect new COMPATIBLE_WITH_RECONFIGURED_SERIALIZER option

Adjust serializer tests to use the reconfigured serializer if such
serializer was provided.

An OptinalMap is a map that allows null values and null keys (as long
as a name is assosicated with them. This would replace the usage of
DummyUnloadableClass and DummyUnloadableSerializer within
KryoSerializerSnapshot.

The test enum type written in test resource files has different
serialVersionUIDs generated between Scala 2.11 / 2.12. The test fails in
2.12 because we wrote the test files with 2.11.

This commit fixes that by explicitly setting a fixed serialVersionUID
for the test enum type.

First, this inverts the if statements to make the code more readable
and maintainable.

Then, this changes the check to report *incompatible* when the order
of enums changes or when enums are removed because we currently don't
maintain a map of enum ids to enum values ourselves and therefore cannot
migrate or reconfigure the serializer to read data that was serialized
with the old mapping.

Before, this was trying to create an Enumeration and cast it to a
TypeSerializer. Creating the enumeration would fail because it is a
Scala Object, i.e. a singleton. The cast would also fail.

This now uses inspection to create an instance of the Enumeration and
then uses that to create an EnumValueSerializer.

This logic will never be used because the new snapshot will never be
used to deserialize an old config snapshot.

Using an IntSerialize technically makes EnumValueSerializer a composite
serializer, which it doesn't have to be. This decouples
EnumValueSerializer from any future changes to IntSerializer.

This commit migrate EnumValueSerializer to use new serialization compatibilty abstractions
  * add a new class `ScalaEnumSerializerSnapshot`
  * return a `ScalaEnumSerializerConfigSnapshot ` with `ScalaEnumSerializerSnapshot` when calling `EnumValueSerializer#snapshotConfiguration`
  * add a migration test `EnumValueSerializerSnapshotMigrationTest` to test the compatibility
  * remove function `EnumValueSerializer#ensureCompatibility()`

This closes #7753.

[FLINK-11699] Use isEmpty() for address comparison in AbstractDispatcherResourceManagerComponentFactory

This closes #7783.

We also don't check the state on every call anymore.

This makes it easier to read messages of failing tests.

This makes it easier to understand failing tests where serializers
are being compared.

Before, it wasn't comparing the ordering of enum values, i.e. the
mapping of indices to enum values.

This changes the new EnumSerializer Snapshot to hold Enums, instead of
their String representation (we to the mapping from and to Enum in the
read/write methods).

This also changes the compatibility checks to detect changes in enum
ordering and then report that only reconfiguration is required.