__buffa_cached_size makes destructuring harder

[Guiguiprim]

In some cases it is useful to de-structure a message (for example to be sure not to forget to add a handling in case of field addition).

The presence of __buffa_cached_size in all messages makes it more painful.

Does the size really need to be cache inside the struct, can't it be done outside where it is needed ?

[iainmcgin]

We can remove the cached size from the structs if we produce a post-order Vec<u32> during encoding as ephemeral state, but there's a related issue with __buffa_unknown_fields, which contains real state from decoding that is required for unknown field preservation (which is optional). I'll try implementing the change to eliminate __buffa_cached_size, but will need to think more on what to do with preserved unknown field data; perhaps __buffa_unknown_fields: _ in the destructuring is an acceptable compromise for that specific case since it has more justification.

[Guiguiprim]

Even more __buffa_unknown_fields can be opt-out, so that is not as much an issue. ( But it could be useful to be able to opt-in back for some specific structures, don't you think ?)