Make Plack::Util::header_* less undertested

[ap]

I am not finished with this branch – there are lots of tests still missing. But I am posting this as an RFC.

1. Do you agree with the split into separate test files?

2. Is the added build-time dependency on Test::Differences OK with you?

   Its eq_or_diff produces much nicer output on test failure than is_deeply does. The downside is that it adds four more transitive dependencies. Fortunately they are all pure-Perl, at least.

   If you don’t want it, I can just s/eq_or_diff/is_deeply/g it out – they are signature-compatible with each other. That means compromise is also possible, i.e. attempting to load Test::Differences and then using either eq_or_diff or is_deeply depending on whether it’s available.

   What is your preference?

[coveralls]

Coverage remained the same at 81.84% when pulling 6dbe592 on ap:header-utils-tests into db4148f on plack:master.

[coveralls]

Coverage remained the same at 81.84% when pulling 6dbe592 on ap:header-utils-tests into db4148f on plack:master.

[coveralls]

Coverage increased (+0.04%) to 81.88% when pulling 6dbe592 on ap:header-utils-tests into db4148f on plack:master.

[miyagawa]

Let's not add Test::Difference - nice output would be only nice when the
test fails :)