Metadata-parser

* added `fix_unicode_url()`
* Added `allow_unicode_url` (default True) to the following calls:
`MetadataParser.get_url_canonical`
`MetadataParser.get_url_opengraph`
`MetadataParser.get_discrete_url`
This functionality will try to recode canonical urls with unicode data into percent-encoded streams

* Python3 support returned

* added some more tests to ensure encoding detected correctly
* stash the soup sooner when parsing, to aid in debugging

* doing some work to guess encoding...
* internal: now using `resp` instead of `r`, it is easier for pdb debugging
* the peername check was changed to be a hook, so it can be processed more immediately
* the custom session redirect test was altered
* changed the DummyResponse encoding fallback to `ENCODING_FALLBACK` which is Latin (not utf8)
this is somewhat backwards incompatible with this library, but maintains compatibility with the underlying `requests` library

* added more attributes to DummyResponse:
** `content`
** `headers`

* some changes to how we handle upgrading bad canonicals
upgrades will no longer happen IF they specify a bad domain.
upgrades from localhost will still transfer over