Use Message.getReaderSchema() in Pulsar IO Sinks when possible

[eolivelli]

Motivation

When you run a Sink and you call record.getSchema(), it does not return an accurate representation of the schema in case of a schema update.

For instance:

• the topic starts with AVRO Schema schema1
• the AutoConsumeSchema starts by populating the internal SchemaInfo with the definition of schema1
• the topic advances to AVRO Schema schema2
• the AutoConsumeSchema still reports SchemaInfo for schema1
• record.getSchema().getNativeSchema() reports the wrong schema definition

Modifications

With this change we leverage PIP-85 Message.getReaderSchema() API that returns the exact schema used for the Message, that is the schema that these pieces for information updated to the same "schemaVersion" of the Message represented by the Record:

• reports the correct getSchemaInfo()
• reports the correct getNativeSchema()

I also fixing a problem in KeyValueSchema for atSchemaVersion(), the constructor of KeyValueSchema did not fill in the SchemaInfo data structure, resulting in NPEs.

Verifying this change

This change adds a new integration test and a unit test

Does this pull request potentially affect one of the following parts:

It affects Pulsar Sinks that implement Sink, in fact now Record.getSchema() will return accurate schema information.

Documentation

No need for docs, previous behaviour was unexpected, the new behaviour is what you expect.

[eolivelli]

I am working on the integration test, I have marked this PR as "draft" for early review

[codelipenghui]

LGTM

[congbobo184]

LGTM!