[tx] Speed up transaction parser

[bgrins]

When importing a map structure via transactions that contains around 1000 lines I'm getting around 3 seconds on the parse

[bgrins]

From slack:

Nice thing is it's trivially parallel
Mmap the file, segment lines, pass offsets to parser, preprocess in parallel, queue for final write

[rnewman]

Though I should add: it's theoretically possible for us to parse this data with zero string copying, and it should be way faster than this, even serially. It might be a lot of work to get there, but it seems wiser to do some of it than to immediately throw cores at the problem!