Some minor optimizations

[travisbrown]

This PR replaces some instances of andThen, compose, etc. with function literals. In all of these cases I think the change improves readability—for example this:

def curry[X, Y, Z](f: (X, Y) => Z): X => Y => Z = x => y => f(x, y)
Kleisli[M, E, A](curry(Function.untupled(f)).andThen(R.tabulate))

Is now this:

Kleisli[M, E, A](e => R.tabulate(r => f((e, r))))

It also improves performance, especially in cases like parBitraverse where just rewriting:

f.andThen(P.parallel.apply(_)), g.andThen(P.parallel.apply(_))

To:

a => P.parallel.apply(f(a)), b => P.parallel.apply(g(b))

…avoids multiple allocations and some indirection.

I've included a couple of new benchmarks, and some example results for Scala 2.12 and JVM 8 on my desktop machine:

Benchmark                                       Mode  Cnt         Score        Error  Units
ParTraverseBench.eitherParBitraversePointfree  thrpt   50  20201469.796 ± 143402.778  ops/s
ParTraverseBench.eitherParBitraversePointfull  thrpt   50  24742265.071 ± 133253.733  ops/s
ParTraverseBench.eitherParTraversePointfree    thrpt   50   1150877.660 ±  10162.432  ops/s
ParTraverseBench.eitherParTraversePointfull    thrpt   50   1221809.128 ±   9997.474  ops/s

[codecov-io]

Codecov Report

Merging #3159 into master will not change coverage.
The diff coverage is 92.3%.

@@           Coverage Diff           @@
##           master    #3159   +/-   ##
=======================================
  Coverage   93.04%   93.04%           
=======================================
  Files         376      376           
  Lines        7374     7374           
  Branches      209      207    -2     
=======================================
  Hits         6861     6861           
  Misses        513      513

Flag	Coverage Δ
#scala_version_212	93.34% <92.3%> (-0.03%)	⬇️
#scala_version_213	90.61% <88.46%> (ø)	⬆️

Impacted Files	Coverage Δ
core/src/main/scala/cats/data/NonEmptyChain.scala	86.02% <0%> (ø)	⬆️
core/src/main/scala/cats/data/Func.scala	96.42% <0%> (ø)	⬆️
core/src/main/scala/cats/data/Validated.scala	100% <100%> (ø)	⬆️
core/src/main/scala/cats/data/NonEmptyList.scala	98.7% <100%> (ø)	⬆️
core/src/main/scala/cats/data/Nested.scala	95.12% <100%> (ø)	⬆️
core/src/main/scala/cats/data/Kleisli.scala	90.82% <100%> (-0.09%)	⬇️
core/src/main/scala/cats/Parallel.scala	84.84% <100%> (ø)	⬆️
core/src/main/scala/cats/data/IdT.scala	97.72% <100%> (ø)	⬆️
core/src/main/scala/cats/data/Cokleisli.scala	100% <100%> (ø)	⬆️
core/src/main/scala/cats/data/NonEmptySet.scala	97.64% <100%> (ø)	⬆️
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8867e97...19a02e1. Read the comment docs.

[rossabaker]

I find most of them lateral in readability, but tabulate is much nicer, and lateral-but-faster is a win. 👍