Add support for skippable named writeables

[ywelsch]

While XContent parsing allows to skip over unknown fields or subobjects, there is no such feature in the StreamInput / StreamOutput world. This PR adds support for skippable named writeables:

• the NamedWriteableRegistry has an additional parameter to specify which categories are allowed to be skipped. For the transport client, we mark ClusterState.Custom and MetaData.Custom as skippable.
• StreamOut has a new method writeSkippableNamedWriteable which calculates the number of bytes to be written for the NamedWriteable and writes this as a header.
• StreamInput has a new method readSkippableNamedWriteable which uses the header written by writeSkippableNamedWriteable to possibly skip the NamedWriteable object.
• ClusterState.Custom and MetaData.Custom are now written and read using these two new methods writeSkippableNamedWriteable and readSkippableNamedWriteable.
• Leniency for skipping the customs is only introduced in the transport client. This allows a transport client to deserialize a cluster state even if does not know about all the customs objects.

[elasticmachine]

Pinging @elastic/es-distributed

[ywelsch]

Closed in favor of #31020