patching #57, $uri->canonical unconditionally returns a clone

[doriantaylor]

This change makes $uri->canonical uniformly return a cloned object. Erstwhile it returned $self if there was nothing to change, for the stated purpose of efficiency, although benchmarks suggest this turns out to be only about a 7% gain on an operation that is measured in microseconds, and only in the no-op scenario.

[karenetheridge]

@doriantaylor can you squash these commits into one, and add "closes #57" to the commit message?

[doriantaylor]

All requested changes made. Worth noting re $ref1 == $ref2 is all over the place internally, so if there's an overload on numeric == then those instances should probably be changed to refaddr too.

[oalders]

@karenetheridge there have been some changes. Can you review them?

[doriantaylor]

are these changes satisfactory?

[oalders]

@doriantaylor not sure if @karenetheridge saw this. I'll touch base with her tomorrow so we can see about getting this merged.

[oalders]

Thanks @doriantaylor and @karenetheridge!

[oalders]

@doriantaylor I just reverted this change for the time being. Please see libwww-perl/HTTP-Message#121

Since this was breaking HTTP::Message installs I figured it would be best to back out of the changes for the time being until we know what the issue with HTTP::Config. Are you able to look into this?

[doriantaylor]

Even money on the test being the problem. I'll check it out.

[doriantaylor]

eh spoke too soon; looking at matching_items