Revisit String performance

[anthony-swirldslabs]

UTF-8 encoding takes time. Many times, we don't even use strings that we parse (e.g. memo fields and similar). So PBJ may be wasting time encoding them when parsing objects.

The idea is to store strings as original byte[] objects in models and only perform the encoding if/when someone calls a getter on the model for the string. If the getter is never called, then we never need to waste cycles on the encoding.

A PoC of this change is available at #612

The task is to extract it, get it to the production level, and release the change.

[anthony-swirldslabs]

The #636 has been tested for performance, and the results look like this:

In other words, the performance is on par in many benchmarks (meaning we read the string fields once, so we don't win anything), or 10% worse in case of smart contracts (likely because we read the string fields multiple times and hence lose).

So we're parking the story for now because the optimization doesn't seem to deliver the desired gains.