when drafting the BIP I attempted to triangulate what the width of the fields for number of inputs/outputs should be from various sources in the codebase, and made an error in looking at the signing and encoding logic, and not the _decoding_ logic which restricts vectors (vin, vout) to MAX_SIZE which is 33554432. This fully justifies not using a wider type (uint64_t).
Also clarified arguments for not using a narrower type (uint16_t) which would be possible just for the vIn based on block size, because it's a leaky abstraction (you can still encode and decode such a transaction, just not mine it).
thanks to @roconnor-blockstream for pointing this out
