perf: effective data model

[ernado]

Implement (internal) data model representation to reduce allocations and CPU overhead.

This model should be effectively (un)-marshaled into binary representation so it can be also used on-the-wire, e.g. for sending via message broker like Kafka.

Advanced optimizations like string interning, memory re-use, pooling, etc are welcome.

[ernado]

Also model should be optimal for https://github.com/ClickHouse/ch-go/tree/main/proto serialization