Use C size for serialization/deserialization

[lucic71]

This adds a new method in ByteRepr called byte_size. It is used to describe the C/C++ byte size of the object and it's used for serializatoin/deserialization through from/to_bytes.

In refcount, Rust structs have a different size than the C/C++ structs because each field is wrapped in Value<>. For example:

struct S {
  int32_t a;
  int32_t b;
}
sizeof(S) == 8

becomes

struct S {
  a: Value<i32>,
  b: Value<i32>
}
size_of(S) == 16 // one Rc is 8

This difference in size breaks ByteRepr for structs. To do correct serialization we need the C/C++ size. Codegen embeds this into the generated code using the ByteRepr::byte_size method.

Then, when a reinterpreted pointer is created, ReinterpretedView::elem_byte_size saves ByteRepr::byte_size. This is an optimization so that the ByteRepr trait is not exposed to all Ptr methods, even into methods that don't have to do with reinterpreted pointers.

The goal of this is to support structs inside unions. Structs, as any other type that takes part in an union, is a reinterpreted view into the byte buffer of the union.