fix(pypi): make the URL/filename extraction from requirement more robust (#2871)

Summary:
- Make the requirement line the same as the one that is used in whls. It
  only contains extras and the version if it is present.
- Add debug log statements if we fail to get the version from a direct
  URL reference.
- Move some tests from `parse_requirements_tests` to
  `index_sources_tests` to improve test
  maintenance.
- Replace the URL encoded `+` to a regular `+` in the filename.
- Correctly handle the case when the `=sha256:` is used in the URL.

Once this is merged I plan to tackle #2648 by changing the
`parse_requirements` code to de-duplicate entries returned
by the `parse_requirements` function.

I cannot think of anything else that we can do for this as of now, so
will mark the associated issue as resolved.

Fixes #2363
Work towards #2648