Remove `Mempool.`getTxSize`; the snapshot interface contains sizes now.

Transaction size, block capacity, and mempool capacity are multi-dimensional
vectors (incl ExUnits, etc), instead of merely bytes: see `TxMeasure`.

A transaction cannot be added if it would push any component of the size over
that component's capacity.

The capacity override is still only specified in byte size, but the value is
interpreted as a block count (rounded up).

Enforce block capacity _before_ the forging logic.
Now the forging logic simply includes whatever transactions its given, which is
reasonable and simpler. It's the NodeKernel logic that uses the mempool's
finger tree in order to slice an appropriately-sized prefix, which is then
passed to the now-dumb forging function.

consensus: no longer need BoundedMeasure, Measure suffices

consensus-cardano: use ConwayMeasure in Babbage

TOSQUASH fixup overflow workarounds and their explanations

consensus: update stale tryAddTxs Haddock references

consensus: an editing pass on the addTx Haddock

consensus: elaborate comment on pureTryAddTx guard

consensus: mitigate new mempool DoS vector

consensus: txMeasure now fails if the tx is too big

TOSQUASH fixup hlint and 9.* build

TOSQUASH fixup stylish

TOSQUASH polishing

consensus: explicit overflow check in pureTryAddTx

consensus: Word32 for tx byte size instead of Natural

consensus: mention Segregated Witness scheme in tx byte size comment

…e name

…will likely explode due to overflow

… in network