* Various microoptimisations to knit and gzip - reducing function call
counts in inner loops, reducing double-handling of data, increasing
knit header parsing performance, removing 'list.index' calls
completely etc.
* changes the data format of knits to remove an extraneous \n in the
compressed content data, which lets it be trivially pulled into
memory in a single 'readlines()' call.
* changes the knit repository to use unannotated knits for inventory,
revisions, and signatures. Also turns off delta-compression for
revisions and signatures.