Mempool test: generate txs larger than the entire mempool

[amesgen]

Closes #1226

In addition to the previously valid/invalid txs (purely based on the UTxO ledger rules), we add an optional per-tx limit to the mock block.

As a second step, we generate very large txs that are larger than an entire mempool, in order to test that we do not block when adding them (just like the other txs), which is important as explained in #1226.

---

One way to validate this PR is to introduce a bug that would cause us to block on such transactions, and observe that the test now indeed catches that.

For example, retrying when the per-tx limited is not satisfied (this is basically what happened in #1168 and was fixed by #1225) via

diff --git a/ouroboros-consensus/src/ouroboros-consensus/Ouroboros/Consensus/Mempool/Update.hs b/ouroboros-consensus/src/ouroboros-consensus/Ouroboros/Consensus/Mempool/Update.hs
index 372ea15c29..31abe25fbf 100644
--- a/ouroboros-consensus/src/ouroboros-consensus/Ouroboros/Consensus/Mempool/Update.hs
+++ b/ouroboros-consensus/src/ouroboros-consensus/Ouroboros/Consensus/Mempool/Update.hs
@@ -170,25 +170,8 @@ pureTryAddTx ::
   -> TryAddTx blk
 pureTryAddTx cfg wti tx is =
   case runExcept $ txMeasure cfg (isLedgerState is) tx of
-    Left err ->
-      -- The transaction does not have a valid measure (eg its ExUnits is
-      -- greater than what this ledger state allows for a single transaction).
-      --
-      -- It might seem simpler to remove the failure case from 'txMeasure' and
-      -- simply fully validate the tx before determining whether it'd fit in
-      -- the mempool; that way we could reject invalid txs ASAP. However, for a
-      -- valid tx, we'd pay that validation cost every time the node's
-      -- selection changed, even if the tx wouldn't fit. So it'd very much be
-      -- as if the mempool were effectively over capacity! What's worse, each
-      -- attempt would not be using 'extendVRPrevApplied'.
-      TryAddTx
-        Nothing
-        (MempoolTxRejected tx err)
-        (TraceMempoolRejectedTx
-         tx
-         err
-         (isMempoolSize is)
-        )
+    Left _err ->
+      NotEnoughSpaceLeft
     Right txsz
       -- Check for overflow
       --

or here

diff --git a/ouroboros-consensus/src/unstable-mock-block/Ouroboros/Consensus/Mock/Ledger/Block.hs b/ouroboros-consensus/src/unstable-mock-block/Ouroboros/Consensus/Mock/Ledger/Block.hs
index 743e11bc61..e01d8cfe5a 100644
--- a/ouroboros-consensus/src/unstable-mock-block/Ouroboros/Consensus/Mock/Ledger/Block.hs
+++ b/ouroboros-consensus/src/unstable-mock-block/Ouroboros/Consensus/Mock/Ledger/Block.hs
@@ -452,10 +452,7 @@ instance TxLimits (SimpleBlock c ext) where
   -- But not 'maxbound'!, since the mempool sometimes holds multiple blocks worth.
   blockCapacityTxMeasure _cfg _st = IgnoringOverflow simpleBlockCapacity
 
-  txMeasure cfg _st =
-        fmap IgnoringOverflow
-      . checkTxSize (simpleLedgerMockConfig cfg)
-      . simpleGenTx
+  txMeasure _cfg _st = pure . IgnoringOverflow . txSize . simpleGenTx
 
 simpleBlockCapacity :: ByteSize32
 simpleBlockCapacity = ByteSize32 512

will cause many test failures with FailureDeadlock [Labelled (ThreadId []) (Just "main")]' via io-sim's nice deadlock detection.

---

Stacked on top of #1175 to avoid boring rebase work

[nfrisby]

Just a few minor changes

[nfrisby]

This comment is a bit confusing. Does mustBeValid not check the per-tx size limit?

[amesgen]

It does not as we use testLedgerConfigNoSizeLimits in this particular case. The current approach is to first generate "non-large" txs, and then use that to set the mempool size (already before this PR) and the per-tx limit.