feat: encoder

[william-silversmith]

This PR adds an spng encoder to the library and builds off of its sister PR #6 for improving the build process.

The encoder currently supports regular, progressive, and interlaced modes as well as a compression level setting. No support for palettes or optimization at this time.

import pyspng
from pyspng import ProgressiveMode
binary = pyspng.encode(numpy_array, progressive=ProgressiveMode.NONE, compress_level=6)

This also fixes an issue where spng was not able to decode GRAY+ALPHA colorspaces (see randy408/libspng#207) because the author of this library was so careful as to specify fmt. I replaced the spng calls with SPNG_FMT_PNG to enable that decoding.

The tests were updated to include synthetic encoding tests for correctness and benchmark performance. Here are my results on a Macbook Pro M1:

d: pil  compressed   noise     272.80ms
d: pil  compressed   gradient  140.74ms
d: pil  uncompressed noise     86.95ms
d: pil  uncompressed gradient  82.32ms
d: spng compressed   noise     148.25ms
d: spng compressed   gradient  73.53ms
d: spng uncompressed noise     30.85ms
d: spng uncompressed gradient  33.78ms
e: pil  compressed   noise     1534.95ms
e: pil  compressed   gradient  462.84ms
e: pil  uncompressed noise     285.58ms
e: pil  uncompressed gradient  127.78ms
e: spng compressed   noise     1153.44ms
e: spng compressed   gradient  272.75ms
e: spng uncompressed noise     397.57ms
e: spng uncompressed gradient  398.16ms

I might add a couple more commits tomorrow. Thanks for reading and for developing a great library!