Pure-haskell "memchr"

[chreekat]

This removes use of utils.c under -fpure-haskell.

(Technically there is still one use site of _hs_text_memcmp2, but it's guarded by MIN_VERSION_BASE(4,11,0). Thus -fpure-haskell is not compatible with ghc-8.2. I think I'm ok with that.)

This PR will conflict with #545 , so I'll keep this one a draft until that one is finalized. Plus, my use of toList is probably dumb, and I'd like to come up with a better way of doing it.

[chreekat]

I just noticed that all callsites are of form cSizeToInt . memchr, so I'll probably move that into the definition, just as I did with the arguments.

[chreekat]

What would be some good benchmarks to run? I reached blindly and grabbed these ones:

./dist-pure/build/x86_64-linux/ghc-9.2.8/text-2.1/b/text-benchmarks/build/text-benchmarks/text-benchmarks --baseline fast.baseline -p '$0 ~ /Sort/ || $0 ~ /Indices/ || $0 ~ /Lines/'
All
  ReadLines
    Text:                 OK (0.32s)
      44.8 ms ± 4.5 ms,       same as baseline
  FileIndices
    Text:                 OK (0.25s)
      85.2 ms ± 2.7 ms, 6968% more than baseline
    LazyText:             OK (0.27s)
      2.22 ms ± 123 μs,       same as baseline
  Programs
    Sort
      Text:               OK (0.80s)
        268  ms ± 6.7 ms, 87% more than baseline
      LazyText:           OK (0.89s)
        296  ms ±  13 ms, 58% more than baseline
      TextByteString:     OK (0.59s)
        195  ms ± 5.8 ms, 344% more than baseline
      LazyTextByteString: OK (0.56s)
        186  ms ±  14 ms, 258% more than baseline
      TextBuilder:        OK (0.57s)
        190  ms ± 5.6 ms, 318% more than baseline

(This is comparison of -fpure-haskell versus -fno-pure-haskell)

[Bodigrim]

I'm not too worried about performance in -fpure-haskell mode, so overall looks sane to me as is.

[chreekat]

This is ready now.

[Bodigrim]

Yay! Thanks @chreekat!

[chreekat]

Awesome! Now I can go back to converting my console app to a web app 😂