switch to lz4_flex

[PSeitz]

The release of lz4_flex 0.11 fixes two issue that caused problems
before:

• The new auto-finish flushes on drop
• FrameEncoder writes now empty frames. Previously nothing was written,
  which caused issues with double compressed empty data.

[marcospb19]

Sorry for the delay.

I did some benchmarks before merging.

The more I chain decoders, the better results I get:

1. .lz4 -> 1.3x
2. .lz4.lz4 -> 1.6x
3. .lz4.lz4.lz4.lz4 -> 2.0x

Summary
  ./ouch-new compress input.txt output.lz4.lz4 ran
    1.63 ± 0.02 times faster than ./ouch-previous compress input.txt output.lz4.lz4

I also saw improvements when decompressing.

Input file was generated with the following command:

dd if=/dev/random of=input.txt bs=1M count=500

All tests done in /tmp, because it is a filesystem that lives in-memory (doesn't add disk IO to the mix).

Benchmark command used:

hyperfine -w 3 -p 'rm output.lz4.lz4' './ouch-previous compress input.txt output.lz4.lz4' './ouch-new compress input.txt output.lz4.lz4'

In my machine (using nightly-x86_64-unknown-linux-gnu), I also saw a reduction in binary size of 123kb (from 3589 to 3466) (yes I did rebase this PR to account for other changes).

Thanks for your contribution!

[figsoda]

Thank you for benchmarking this @marcospb19, and thank you for implementing this and fixing the bugs @PSeitz! I didn't have to run the benchmarks initially and kinda forgot about it (sorry), this looks great!