Python-rsync

* Rename the library to `librsync'.

* Portability fixes.

* Include the popt library, and use it to build rdiff if the library
is not present on the host.

* Add file(1) magic for rdiff.

* Add more to the manual pages.

* It's no longer necessary to call rs_buffers_init on a stream before
starting to use it: all the internal data is kept in the job, not
in the stream.

* Rename rs_stream_t to rs_buffers_t, a more obvious name. Pass the
buffers to every rs_job_iter() call, rather than setting it at
startup. Similarly for all the _begin() functions.

* rs_job_new also takes the initial state function.

* Return RS_PARAM_ERROR when library is misused.

* Redesign API to be more like zlib/bzlib.

* Put all command-line functions into a single rdiff(1) program.

* New magic number `rs6'

* Change to using popt for command line parsing.

* Use Doxygen for API documentation.

* Changes stats string format.

* Slightly improved test cases

* Don't install debugging tools into /usr/local/bin; leave them in
the source directory.

* Fix libhsync to build on (sgi-mips, IRIX64, gcc, GNU Make)

* Include README.CVS in tarball

* Back out of using libtool and shared libraries, as it is
unnecessary at this stage, complicates installation and slows down
compilation.

* Use mapptr when reading data to decode, so that decoding should
have less latency and be more reliable.

* Cope better on systems that are missing functions like snprintf.

* Put genuine search encoding back into the nad algorithm, and
further clean up the nad code. Literals are now sent out using a
literal buffer integrated with the input mapptr so that data is not
copied. Checksums are still calculated from scratch each time
rather than by rolling -- this is very slow but simple.

* Reshuffle test cases so that they use files generated by hsmapread,
rather than the source directory. This makes the tests quicker and
more reproducible, hopefully without losing coverage. Further
develop the test driver framework.

* Add hsdumpsums debugging tool.

* Hex strings (eg strong checksums) are broken up by underscores for
readability.

* Stats now go to the log rather than stdout.

* mapptr acts properly when we're skipping/rewinding to data already
present in the buffer -- it does a copy if required, but not
necessarily real IO.

* Improved mapptr input code

* Turn on more warnings if using gcc

* More test cases