Redo benchmarks

[RyanGlScott]

Attempting to build bench:cryptol-bench fails with:

$ cabal build bench:cryptol-bench
Build profile: -w ghc-8.10.7 -O1
In order, the following will be built (use -v for more details):
 - cryptol-2.12.0.99 (bench:cryptol-bench) (first run)
Preprocessing benchmark 'cryptol-bench' for cryptol-2.12.0.99..
Building benchmark 'cryptol-bench' for cryptol-2.12.0.99..
[1 of 1] Compiling Main             ( bench/Main.hs, /home/rscott/Documents/Hacking/Haskell/cryptol/dist-newstyle/build/x86_64-linux/ghc-8.10.7/cryptol-2.12.0.99/b/cryptol-bench/build/cryptol-bench/cryptol-bench-tmp/Main.o )

bench/Main.hs:151:23: error:
    Not in scope: ‘C.evalPrim’
    Module ‘Cryptol.Eval.Concrete’ does not export ‘evalPrim’.
    |
151 |       let ?evalPrim = C.evalPrim
    |                       ^^^^^^^^^^

bench/Main.hs:172:57: error:
    Not in scope: ‘S.sbvEval’
    Module ‘Cryptol.Eval.SBV’ does not export ‘sbvEval’.
    |
172 |     bench name $ whnfIO $ fmap force E.runEval evOpts $ S.sbvEval $ do
    |                                                         ^^^^^^^^^

bench/Main.hs:173:23: error:
    Not in scope: ‘S.evalPrim’
    Module ‘Cryptol.Eval.SBV’ does not export ‘evalPrim’.
    |
173 |       let ?evalPrim = S.evalPrim
    |                       ^^^^^^^^^^

bench/Main.hs:174:27: error:
    Not in scope: data constructor ‘S.SBV’
    Module ‘Cryptol.Eval.SBV’ does not export ‘SBV’.
    |
174 |       env' <- E.evalDecls S.SBV (M.allDeclGroups menv) mempty
    |                           ^^^^^

bench/Main.hs:175:13: error:
    Not in scope: type constructor or class ‘S.Value’
    Perhaps you meant ‘C.Value’ (imported from Cryptol.Eval.Concrete)
    Module ‘Cryptol.Eval.SBV’ does not export ‘Value’.
    |
175 |       (e :: S.Value) <- E.evalExpr S.SBV env' texpr
    |             ^^^^^^^

bench/Main.hs:175:36: error:
    Not in scope: data constructor ‘S.SBV’
    Module ‘Cryptol.Eval.SBV’ does not export ‘SBV’.
    |
175 |       (e :: S.Value) <- E.evalExpr S.SBV env' texpr
    |                                    ^^^^^

It's possible that #928 is the culprit.

[qsctr]

After a conversation with @robdockins it seems like the existing benchmarks have bit-rotted so much that it's probably a better choice to delete them and start over, and they were not really fully achieving what we wanted them to do anyways. But we need to spend some time thinking about what exactly we want to measure, what is the best way to measure it, and how it would fit into CI (e.g. would we make CI fail on a performance regression?).