Don't log a warning on content-type: multipart/* responses

[badcure]

This is to address the multipart errors that are showing up in urllib. The real fix is upstream, but the goal is to remove the false warnings that are showing up as described in #800 .

If the exceptions show up, I am simply removing them from the list. The exceptions would only apply if it is processing the body as well. In the urllib3 code path, it only cares about the headers at that point.

Closes #800

[codecov-io]

Codecov Report

Merging #1665 into master will decrease coverage by 0.4%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #1665      +/-   ##
==========================================
- Coverage   99.75%   99.35%   -0.41%     
==========================================
  Files          22       22              
  Lines        2006     2000       -6     
==========================================
- Hits         2001     1987      -14     
- Misses          5       13       +8

Impacted Files	Coverage Δ
src/urllib3/util/response.py	100% <100%> (ø)	⬆️
src/urllib3/connection.py	93.75% <0%> (-4.2%)	⬇️
src/urllib3/response.py	99.44% <0%> (-0.56%)	⬇️
src/urllib3/util/url.py	98.97% <0%> (-0.03%)	⬇️
src/urllib3/util/timeout.py	100% <0%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d369183...18fd8a8. Read the comment docs.

[badcure]

Hello @pquentin , wanted to see if you had time to look over this. The goal of this is to remove the warnings, and wanted to see if there were concerns going down this route.

[pquentin]

@badcure Thank you for putting this pull request together! It's easy to read, and the change does make sense.

However, I don't know enough about header parsing to reason about possible issues (eg. security) so I'm not going to merge this myself, sorry. :(

[harryhobson]

Hi @pquentin, who is the right person to review and accept this pull request? It has been sitting dormant for a while so I just want to check this hasn't been forgotten.

[pquentin]

@sethmlarson Any ideas?

[sethmlarson]

Sorry that this took so long, I've done the following:

• Reviewed the change, I believe that this is a good change for us 👍
• Confirmed that totally valid HTTP multipart/mixed responses receive these defects (even when using raw httplib) so we definitely should be ignoring them if we're only parsing headers.
• Added a test case that asserts that the log.warning() isn't called when receiving a multipart/mixed message
• Made some changes to the PR to remove the unnecessary pytest.mark.parametrize() and use onlyPy3
• Reworded the docs for the behavior and simplified the filtering

@pquentin @hodbn could one of you review my changes to this PR?