// TODO for any specific type, we should be able to tersely create a new type that embeds its assembler and wraps the one method that's valid for finishing its kind.
// yes, do lets use unexported 'func()' callbacks for finishers, and thus generate WAY less types.
// adding a word of memory to every builder? not really. the childbuilders would anyway for '*p' instead. toplevels grow, true; but they frankly don't matter: 'one' doesn't show up on an asymtote.
// is it an indirect call that can't be inlined? yeah. but... good lord; there's enough of them in the area already. this one won't tip anything over.
// one per field anyway? probably, at least on the first pass. easier to write; minimal consequence; can optimize size later.
// '_MaybeT__Assembler'? probably, yes -- the 'w' pointer has to be a different type, or things would get dumb. ugh, every valid assign method is near-dupe but not quite. shikata ga nai?
// will there be '_MaybeT__ReprAssembler'? probably, yes -- how would there not be?
// Wild alternative: every '_T__Assembler' has a 'null bool' in addition to the 'finishCb func()'. If used a context where null isn't valid: check it during finish.
// Scalar builders used as a root can do this via an overriden finisher method (literally, the AssignNull) on the Builder; easy peasy.
// Child assemblers can do it during the 'finishCb'.
// It's important to still return the possible rejection of null in the AssignNull method.
// But I think this shakes out: finishCb gets called at the end of *any* finisher method -- **that includes AssignNull**.
// If you're handling a MaybeT, the 'finishCb' can set the null state even though the '_T__Assembler' can't see the '_MaybeT.m'!
// Similarly: '_T__ReprAssembler' sprouts 'z bool' and 'finishCb func()' and does double duty.
