Improve extract performance via ignoring directories early during os.walk

[akx]

Closes #694 (rebases and reimplements it)

[codecov]

Codecov Report

Attention: Patch coverage is 92.85714% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 90.87%. Comparing base (e0d1018) to head (fc8fca4).
Report is 1 commits behind head on master.

❗ Current head fc8fca4 differs from pull request most recent head b5c2f1c. Consider uploading reports for the commit b5c2f1c to get more accurate results

Files	Patch %	Lines
babel/messages/extract.py	92.85%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #968      +/-   ##
==========================================
- Coverage   90.99%   90.87%   -0.13%     
==========================================
  Files          26       25       -1     
  Lines        4444     4341     -103     
==========================================
- Hits         4044     3945      -99     
+ Misses        400      396       -4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[akx]

@DenverCoder1 If you have a moment, could you review this? :)

[DenverCoder1]

Looks good to me (I haven't tested)

I noticed a similar method in frontend.py -

babel/babel/messages/frontend.py

Lines 277 to 288 in 373a52f

	def _make_directory_filter(ignore_patterns):
	"""
	Build a directory_filter function based on a list of ignore patterns.
	"""
	def cli_directory_filter(dirname):
	basename = os.path.basename(dirname)
	return not any(
	fnmatch.fnmatch(basename, ignore_pattern)
	for ignore_pattern
	in ignore_patterns
	)
	return cli_directory_filter

- it seems to not need to create relative paths there, but maybe it's different. Possibly some logic could be combined or renamed to be consistent, but besides that, it seems like this should work.