* Add test that reproduces the issue

* Suggested fix - always save data in sigterm exit flow

* Address test failures on MacOS due to lack of 'Terminated' output on SIGTERM

On the coverage.py metacov, this makes the HTML 5% smaller.

* Put contexts dict in <script> and reveal them using JavaScript

Fixes #1584

* Address review comments

* coverage/htmlfiles/style.css: Align the contexts to the right

* coverage/htmlfiles/style.scss: Match style.css

* coverage/html.py: Fix linting and mypy errors

* two_tests_py.html: make update-gold

And make it easier to add new gold tests.

There is logic to not measure third-party packages inside configured sources. However, when a (i.e. another) configured source was inside a third-party location, this logic was previously disabled completely.

This caused a problem if a virtual env is set up inside a configured source directory and a configured source package gets installed inside the virtual env. Previously in this case, coverage was measured for all files in the virtual env for the reason described in the previous paragraph.

This commit changes the code to collect all configured source directories inside third-party locations and disable coverage for code in third-party locations only if its not in one of these collected source directories.

* fix: generate xml report packages correctly on windows

* test: check duplicate package names in xml report

* fix: shorten long line in test_xml