Improve performance of computation of sizeTxF for ShelleyTx and AlonzoTx.

[lehins]

There should be no reason why ShelleyTx retains the whole of original bytes, because it can't do so in a block anyways. The only reason why it uses MemoBytes is for size calculation needed for sizedTxF. Remove usage of MemoBytes for ShelleyTx, thus avoiding keeping around bytes for . However in order for us to memoize the size needed for computing the fee we should add a field to AlonzoTx and ShelleyTx. This will require creation of AlonzoTxRaw and removal of MemoBytes for ShelleyTx

[lehins]

Giving it a bit more thought, I think current way of computing the size of transaction for the purpose of fee calculation is unnecessarily very expensive. Current approach combines parts of transaction together as a bytestring and then the size of the bytestring is computed.
We can instead just get the sizes of components and add the constant overhead of the combined version. This approach would allow us to avoid this unnecessary construction of a bytestring just to get the size.

This means that we don't really need to add an extra field anymore, because that approach described above is very cheap and the result doesn't need to be memoized