Fix label recalculation (codecov#200)

When Codecov doesn't return the absent labels we were re-calculating only the absent labels.
The issue is that, becaue Codecov didn't have them in the first place they couldn't filter out labels that might have been changed in the current commit. And we (CLI) weren't doing that either.

That's an issue because pytest fails if you tell it to run a label that doesn't exist.
So if we get the payload back with no absent labels we have to recalculate everything.

Merge pull request codecov#203 from codecov/scott/increment-version

Increment version to v0.1.15

bump version (codecov#198)

bump version to create next release (codecov#180)

Merge pull request codecov#177 from codecov/dana/v0.1.12

upgrade version to 0.1.12

Merge pull request codecov#167 from codecov/dana/0.1.11-release

make a new release

Update setup.py (codecov#157)

Prepare next release

Bump release number to v0.1.9 (codecov#151)

Bump version number (codecov#145)

Encode PUT data before sending POST request to Codecov (codecov#144)

When testing ATS in the sentry fork I have 3 out of 5 uploads timing out when pushing to storage.

I noticed that we are generating the PUT payload *after* receiving the response from Codecov.
Generating PUT payload includes compressing the report file, that might be pretty big,
and considering that the pre-signed put URL has a TTL of 10s this time might be influential.

So now we are just prepping the data completely before starting the upload process,
in the hopes that this will save us enough time to get all uploads into storage in time.