Implement 'unlines' directly

[clyring]

Closes #478.

I wanted to share some code between strict unlines and strict intercalate, but wasn't able to easily get the singleton-copying to rewrite to a poke or similar, resulting in a performance hit. I think this level of duplication is acceptable.

My idea to make Lazy.concat fuse with the existing lazy unlines definition was a slight improvement, but just writing the fused fold manually produced bigger gains. I think this is because now no attempt is made to iterate over a singleton.

Benchmark results:

master:

  unlines
    lazy:   OK (0.94s)
      854  μs ±  17 μs
    strict: OK (0.74s)
      688  μs ±  13 μs

topic branch:

  unlines
    lazy:   OK (19.72s)
      599  μs ± 4.7 μs
    strict: OK (68.04s)
      249  μs ±  18 μs

[sjakobi]

The code looks good!

But "near-inverse" sounds pretty mysterious. It would be good to clarify what that means (an example would be helpful), or alternatively drop that word entirely.

[clyring]

A digression into the ways in which lines and unlines fail to be perfect inverses would be out-of-place, I think. I'd go for left- or right-inverse, if either one was quite correct. There's a restriction on both domain and co-domain that works, but it's a bit wordy to describe directly. There is a mathematical notion called inverse that does apply, but it's not the first one most people think of when they read the word without more context.

[Bodigrim]

Thanks @clyring!