Benchmark and optimise reward calculation

[nc6]

This PR adds a micro-benchmark for the reward update creation and subsequently optimises that operation.

Along the way, we do a few things:

• Make some formatting changes (mostly splitting very long lines)
• Add an ability to plug in a custom TX generator to the block generator (we use this to generate empty blocks).
• Add a function in the benchmarks package to generate a long chain of empty blocks, with delegation initiated at genesis. This moves one function from ouroboros-consensus into shelley-spec-ledger-test.
• Adds two benchmarks:
  • One for createRUpd, built upon a generated chain with some praos blocks issued.
  • One for likelihood, which was the majority of the reward calculation
• Optimise the likelihood calculation. This brings the micro-benchmark down from 50-60ms to around 1.

[nc6]

Note that after this PR the createRUpd function is dominated by this epression:

 Stake $ eval (dom (delegs ▷ Set.singleton hk) ◁ stake)

This already has an optimisation in SetAlgebraInternal:

 -- This case inspired by set expression in EpochBoundary.hs
compute (DRestrict (Dom (RRestrict (Base MapR delegs) (SetSingleton hk))) (Base MapR state)) =
   materialize MapR (do { (x,y,z) <- delegs `domEq` state; when (not (y==hk)); one(x,z) })

So unless @TimSheard spots anything that could be improved there, this is probably in a decent state.

[JaredCorduan]

looks good to me, thank you @nc6 !

@TimSheard does have some ideas for how to make it even faster.

[TimSheard]

This looks good, but Nicks comment about the bottleneck, reminded me about another optimization for a similiar SetAlgebra expression found in LedgerState.hs. This was made considerably faster by writing a function over Data.Maps directy. This function was called intersecDomP, I wrote a simliar function intersectDomPLeft, that applie the same kind of improvements.

I wrote some property tests that test that both functions meet their specifcation.
Benchmarking lead to another improvement where we normailize all singleton sets into one kind of Exp. This allows fewer pattern matches.

Nicks Benchmark rewards/createRUpd varys greatly from run to run (lowest 24 ms greatest 928 ms), Swapping in interesectdomPLeft, also varies greatly but across 10 runs, consistently averages 2-3 times faster.

[nc6]

@TimSheard That last commit seems to cause the tests to break; since we wanted to have this in the release today, I've split it into a separate PR (#1857)

[dcoutts]

LGTM