Experimental support for LZMA / XZ codec

[milesgranger]

Part of #126

Adds experimental lzma / xz support under the experimental module, with a limited amount of configuration, only being able to set preset for compression is all.

67f6902 will close #123 (hopefully) :)

---

For this, it creates a byte-for-byte mirror of the builtin python module using defaults.

In [1]: import lzma

In [2]: import cramjam

In [3]: compressed = lzma.compress(b'bytes')

In [4]: bytes(cramjam.experimental.lzma.compress(b'bytes'))
Out[4]: b'\xfd7zXZ\x00\x00\x04\xe6\xd6\xb4F\x02\x00!\x01\x16\x00\x00\x00t/\xe5\xa3\x01\x00\x04bytes\x00\x00\x00\x006\x93\x11\xb1PA\x11\xab\x00\x01\x1d\x05\xb8-\x80\xaf\x1f\xb6\xf3}\x01\x00\x00\x00\x00\x04YZ'

In [5]: compressed
Out[5]: b'\xfd7zXZ\x00\x00\x04\xe6\xd6\xb4F\x02\x00!\x01\x16\x00\x00\x00t/\xe5\xa3\x01\x00\x04bytes\x00\x00\x00\x006\x93\x11\xb1PA\x11\xab\x00\x01\x1d\x05\xb8-\x80\xaf\x1f\xb6\xf3}\x01\x00\x00\x00\x00\x04YZ'

TODO:

• At least decode support for legacy LZMA format.
• Multi stream support
• Expose more configuration settings?

[lgray]

@milesgranger the present implementation works in our case, thanks! Strangely hinting the output buffer size doesn't seem to bring any performance improvement compared to the python standard library implementation. Perhaps the data I'm testing isn't large enough to see it. Maybe there's a more optimized lzma implementation out there in rust land, but that can come in time.

Still - seems fit to task! Thanks for the snappy response!

Small update: there appears to be a ~10% improvement for our data, when testing with a larger file. Not huge, but I'll take it.

[milesgranger]

Thanks for the feedback @lgray, will be happy to know what you think of follow-ups as time permits. Good there was a bit of improvement, but wasn't expecting anything amazing. I think they both use the same underlying liblzma under the hood.

Probably will benefit from cramjam's de/compress_into functions if you're able to work that out and re-use buffers in your use case.

[lgray]

@milesgranger sure - @ me when you post follow ups. I can test them fairly quickly.

For (de)compress_into I'll have to tinker with the uproot library a bit more, but it's typically organized very sensibly so I should be able to use those methods.