Levenshtein

* some compilation problems with non-gcc were fixed

* median_improve(), a generalized median improving function, was added
* an arbitrary length limitation imposed on greedy median() result was
removed
* out of memory should be handled more gracefully (on systems w/o memory
overcomitting)
* the documentation now passes doctest

* fixed greedy median() for Unicode characters > U+FFFF, it's now usable
with whatever integer type wchar_t happens to be
* added missing MANIFEST
* renamed exported C functions, all public names now have lev_, LEV_ or
Lev prefix; defined lev_byte, lev_wchar, and otherwise santinized
the (still unstable) C interface
* added edit-ops group of functions, with two interfaces: native, useful
for string averaging, and difflib-like for interoperability
* added an example SequenceMatcher-like class StringMatcher

* a segfault in seqratio()/setratio() on invalid input has been fixed
to an exception
* optimized ratio() and distance() (about 20%)
* Levenshtein.h header file was added to make it easier to actually use
it as a C library

* a segfault in setratio() was fixed
* median() handles all empty strings situation more gracefully

* new functions seqratio() and setratio() computing similarity between
string sequences and sets
* Levenshtein optimizations (affects all routines except median())
* all Sequence objects are accepted, not just Lists