[MRG+2] Improve gunzip performance for big files on Python 3

[fbergen]

Copied from discussion on #3270 , partial fix for #2658

Problem
gunzip method runs forever for big files in python 3

bytes in Python 3 does not include string concatenation improvements that Python 2 has, effectively making the following take forever:

%%time
out = b""
for x in range(10000):
    out += b"." * 8000

Solution
Improve the gunzip method to not copy the entire output string in each iteration, but instead append to list + join, significantly improving performance from O(n^2)

Thanks @kmike for your help debugging!

[kmike]

//cc @nctl144 and @lopuhin, as you're working on Scrapy performance improvements and becnhamarks. This change fixes an issue explained here: #2658 (comment) - do you think large and gzip-compressed responses should be added to https://github.com/scrapy/scrapy-bench? Maybe add an option to reproduce not only HTML content of real-world websites from Alexa top, but raw binary responses and HTTP headers as well?

[kmike]

Looks great, thanks @fbergen! +1 to merge after tests pass.

[codecov]

Codecov Report

Merging #3281 into master will not change coverage.
The diff coverage is 100%.

@@           Coverage Diff           @@
##           master    #3281   +/-   ##
=======================================
  Coverage   82.14%   82.14%           
=======================================
  Files         228      228           
  Lines        9599     9599           
  Branches     1385     1385           
=======================================
  Hits         7885     7885           
  Misses       1456     1456           
  Partials      258      258

Impacted Files	Coverage Δ
scrapy/utils/gz.py	100% <100%> (ø)	⬆️

[cathalgarvey]

Looks good to me, thanks @fbergen!