measureRecord() is inefficient

[OlegMazurov]

Problem

measureRecord() is used in sizing and writing protobuf objects. On complex objects, the method is invoked recursively via a lambda, which incurs CPU and allocation overhead. For message lists, additional allocation overhead comes from implicit iterators (see ProtoWriterTools.sizeOfMessageList()). The necessity to measure an object before allocating a buffer for its serialization doubles the cost of measureRecord() from its explicit and implicit invocations.

The current implementation becomes a major factor in performance impacts of new features in hedera-services.

Solution

Reconsider the current design and implementation of Codec.measureRecord().

Alternatives

No response

[anthony-swirldslabs]

This issue has been fixed at #342 where lambdas were replaced with passing the entire Codec instead.

Also, the implicit iterators were replaced with simple for() loops.