profile *construction* of the parsers

[masklinn]

ua-parser/uap-python#253 indirectly reveals that constructing the parsers is much more expensive here than in re2, a difference factor of 4 was measured between the regex-based parser and the re2-based parser.

Parser construction performance is of a lower interest / relevance than actual resolution as I've been assuming parsers are constructed very rarely, but there might be low hanging fruits a bench would reveal. Although I would not be surprised if it was regex having high construction costs either.