Make CLB back-end fast enough for fuzzing

[uhbif19]

Now one Tx sending for Registry takes around 0.1 seconds. This is way to much for fuzzing, we need numbers like 0.01 or 0.001.

Profiling shows that 25% is taken up mkShelleyGlobals in CLB (can be cached), 40% is for makeTransactionBodyAutoBalance and 21% is taken for applySTSOptsEither (real Ledger work) .

So, theoretically around 80% may be dropped with CLB. This might be okay for small and slow fuzzing about 100-1000 examples. For more we probably should use custom Ledger emulator instead of CLB. Seems like applySTS slow performance seem to be real, it does not seem to be only covered by Plutus (?%) or crypto (2.4%) .

TODO:

• Profile CLB performance
• Check CLB performance is improved by -O
• Support without calling makeTransactionBodyAutoBalance and not enforcing Tx fee
• Decide if to create custom Emulator for such cases of CLB

FYI @euonymos

[uhbif19]

Actually real QD works in reasonable time, I am not sure why. >3k actions are done in ~100 seconds.