[Draft] AVRO module - #208

… serdes and correct schema resolution.

Added a dedicated kafka-avro module providing Confluent Schema Registry serializers and deserializers for both SpecificRecord and GenericRecord, and moved all Kafka+Avro code out of the base kafka module, which no longer depends on Avro or Confluent. Deserialization now resolves the writer schema from the registry by the schema id embedded in the payload and performs Avro writer-to-reader resolution, so decoding stays correct across schema evolution. Avro serdes, registry configuration, and telemetry are wired through the new KafkaAvroModule.

- Fixed(kafka-avro): the deserializer resolves the writer schema from Schema Registry by id and applies writer-to-reader schema resolution; previously the id was ignored and payloads were decoded with the reader schema, producing wrong results once the producer schema evolved.
- Added(kafka-avro): SpecificRecord and GenericRecord serializers/deserializers in the Confluent wire format with per-subject schema-id caching, configurable subject-name strategy (topic, record, topic-record) and auto-register via KafkaAvroSerializerConfig.
- Added(kafka-schema-registry): typed basic-auth and bearer-auth configuration for the Schema Registry client, merged into the client properties.
- Added(avro): standard logical-type conversions (decimal, uuid, date, time, timestamp, local-timestamp) on the generic reader/writer path, so logical types round-trip as their Java representations.
- Added(kafka-avro): Avro registry telemetry for lookup latency, schema-cache hit/miss, and serde errors, backed by Micrometer through a factory/impl telemetry structure and falling back to no-op when no MeterRegistry is present.
- Improved(avro): Breaking: AvroReader exposes only unchecked read methods throwing UncheckedIOException, adds getSchema() and read(Schema writerSchema, ...), and caches a datum reader per writer schema.
- Improved(kafka-ap-ksp): the generated Kafka publisher now calls Serializer.configure() with the driver properties and the key/value flag, so subject naming and auto-register settings take effect.
- Removed(kafka-avro): the @avro tag requirement on AvroReader and AvroWriter injection points, since the type is already self-describing; @avro is kept on the ambiguous Serializer and Deserializer beans.

The change restores a coherent Avro integration around the Confluent wire format, isolates the Kafka+Avro concern in its own module, and closes the schema-evolution correctness gap that made the previous serdes unsafe against a Schema Registry.