Currently, in order to serialize an object, the bindings calculate its size; allocate a buffer with the size; serialize.

It would be nice to compare that with a different strategy: grow the buffer when necessary without calculating size at the beginning.

In some cases, the cost of size calculation may be bigger than potential reallocations:
- some messages may be smaller than the default allocation size so skipping calculation is a net win.
- some messages may have such a complex structure that even reallocation is cheaper than walking the structure to calculate the size.

By getting rid of the size calculation stage, we can also easily guarantee that every data field is accessed only once during serialization (instead of twice as we do today).