As discussed previously php-mf2 has custom handling of textContent to
match with what consumers were expecting to get back. This means all
tests around plain text output have a high tendency to fail.

This change addresses that a little by ignoring whitespace differences
within e-* properties.

Travis was showing weird errors in PHP versions before 7.0 about
duplicate name usage. Weird because the Parser class defined for
the test should live in a completely separate namespace.

We mark a single test as incomplete (ie. unimplemented) because the
parser implements a proposed extension to the mf2 specification while
the test suite has tests that exactly match the specification.

Certain builds of PHP seem to drop specific whitespace during the HTML
parsing step. There seems to be no reason for this and the behaviour
has been seen for versions of PHP ranging all the way from 5.6 to 7.3.
The behaviour seems to be sidestepped by providing any supported
parsing option to the loadHTML method. LIBXML_NOWARNING was chosen as
it seemed like it would have the least impact overall.

For a PHP test to surface the behaviour, as well as the test of the
effect of constants please see:
https://gist.github.com/Zegnat/a94489e9b7d5501193e724e336bc6052

Huge thanks to everyone in #indieweb-dev who went on this journey with
me! Especially @cweiske and @Lewiscowles1986 for all the extra
testing, and @gRegorLove for getting the ball rolling with
parsing options.