bpo-43613: Faster implementation of gzip.compress and gzip.decompress

[rhpvorderman]

This also includes the changes from #25011 for bpo-43612. These make more sense in the context of these changes.

Currently gzip.compress and gzip.decompress are implemented with GzipFile. This is a lot of overhead when a simple in memory compression is needed. As shown in the benchmarks below, the overhead is considerable for datasizes below 4096 bytes (which are probably very common targets for in memory compression and decompression).

This PR changes the implementations to compress and decompress in memory.

I compiled python before and after this change with --enable-optimizations to ensure a fair comparison.
I used this script to benchmark:

import gzip
import pathlib
import statistics
import sys
import timeit

DATA=pathlib.Path(sys.argv[1]).read_bytes()

SIZES = [0, 128, 512, 1024, 4096, 8192, 16384]

def benchmark(bench_string, number=1000, repetitions=10):
    for size in SIZES:
        data = DATA[:size]
        compressed_data = gzip.compress(data)
        timeit_kwargs=dict(globals=dict(**locals(), **globals()),
                           number=number)
        results = [timeit.timeit(bench_string, **timeit_kwargs) for _ in range(repetitions)]
        average = statistics.mean(results)
        print(f"Data size {size}: {round(average * (1_000_000 / number),2)} microseconds average")

if __name__ == "__main__":
    print("gzip compression")
    benchmark("gzip.compress(compressed_data)")
    print()
    print("gzip decompression")
    benchmark("gzip.decompress(compressed_data)")

Before:

gzip compression
Data size 0: 7.92 microseconds average
Data size 128: 12.1 microseconds average
Data size 512: 18.45 microseconds average
Data size 1024: 22.41 microseconds average
Data size 4096: 32.51 microseconds average
Data size 8192: 41.03 microseconds average
Data size 16384: 57.99 microseconds average

gzip decompression
Data size 0: 8.99 microseconds average
Data size 128: 10.26 microseconds average
Data size 512: 12.62 microseconds average
Data size 1024: 13.55 microseconds average
Data size 4096: 21.12 microseconds average
Data size 8192: 30.59 microseconds average
Data size 16384: 61.24 microseconds average

After:

gzip compression
Data size 0: 3.68 microseconds average
Data size 128: 7.64 microseconds average
Data size 512: 14.06 microseconds average
Data size 1024: 17.42 microseconds average
Data size 4096: 27.25 microseconds average
Data size 8192: 37.09 microseconds average
Data size 16384: 53.48 microseconds average

gzip decompression
Data size 0: 1.98 microseconds average
Data size 128: 3.74 microseconds average
Data size 512: 5.36 microseconds average
Data size 1024: 6.72 microseconds average
Data size 4096: 14.1 microseconds average
Data size 8192: 23.57 microseconds average
Data size 16384: 52.72 microseconds average

https://bugs.python.org/issue43613

[rhpvorderman]

Thanks for your review and suggestions @ambv! I updated the code.