Improve performance of beans.xml parsing

See my pull request; it was requested there that I create a JIRA for it. Feel free to edit this JIRA issue as appropriate.

Briefly: I was looking for ways to improve the startup time of the Weld SE container if at all possible. I profiled the startup sequence using the NetBeans profiler and while nothing looked bad at all, I found that I could shave about 20ms off of each beans.xml parse, which was the operation that took the most time by (carefully) sharing the SAXParserFactory and SAXParser used to do the parsing. There are probably other micro-optimizations but I figured I'd start here.