XPath performance improvements

[flavorjones]

From comments at #3378:

• Enable the XPath cache. Nick says "The cache speeds up allocation of XPath objects. In general, it's a good idea to enable it."
• Enable xmlXPathOrderDocElems "which speeds up sorting of node sets."

We should benchmark these to understand why/when to enable them. Should they be on for all xpath queries, or should we let users enable them via a method parameter?

[flavorjones]

I've got a dev branch with a WIP commit implementing these two things, and it's a significant performance boost on nokolexbor's benchmark -- a 2x speedup on top of Nick's xpath optimizations branch.

I'm going to wait for Nick's xpath optimizations to land upstream and then come back and finish this. Pretty exciting!