nurllib.parse

This is a rewrite of large portions of urllib.parse.

What's changed:

• All the parsing is done with regexes, strictly compliant to RFCs 3986, 3987, and 6874.
• Performance is worse. (~2x worse with caching, ~5x worse without caching)
• All deprecated and undocumented components are removed.
• There are new functions that allow you to parse only URIs, IRIs, relative-refs, or irelative-refs.
• NFKC normalization is no longer applied before parsing. The standard recommends that this be the IRI producer's responsibility to do this, not ours.
• Hosts and schemes are normalized to lowercase.
• Percent-encoded bytes are normalized to uppercase.
• urlsplit no longer strips garbage bytes from the beginning and end of its input.
• The Result types can no longer be constructed explicitly.
• URLs are now joined according to the procedure from RFC 3986. This means more path normalization and less fragment normalization than urllib.parse.urljoin.
• Leading 0s are now stripped from port numbers.

What's exactly the same:

• urlencode
• non_hierarchical
• scheme_chars
• parse_qs
• parse_qsl
• quote
• quote_from_bytes
• quote_plus
• unquote_to_bytes
• uses_fragment
• uses_netloc
• uses_params
• uses_query
• uses_relative

What's functionally the same:

Everything else, hopefully

Running the tests

Run ./get_tests.sh, then run python3 test_urlparse.py. A few tests fail, but these are due to the documented changes above.