This makes a few small changes to the original patch so it can compile:
 * fix some include statements; argyll flattened the source layout
 * fix comment format at src/yajl_gen.c:184

longs are 64bit wide, at least on eGLibc/Linux/AMD64, therefore 2^31 is
not an integer overflow.

I've removed the integer overflow because the exact limit is machine
dependent and as far as I've seen the RFC doesn't make any assumptions
towards minimum/maximum width of integers.

implement a more efficient custom state stack (rather than re-using buffers), for a ~10% perf improvement during parsing (more function call overhead savings)

inline a couple state management routines for 12% lex perf improvement - also document a couple more opportunities

inline readChar and unreadChar for marginal improvements in parsing performance (15% for json_verify - which just throws away strings after reading))

fflush output in yajl_test to attempt to get deterministic output ordering across systems (msys+win32 being the problem child)

 * bump yajl version to 1.0.0
 * update TODO list, removing completed tasks
 * update yajl_test to override allocation routines and count allocations/frees, providing an automated standalone way to validate we're not leaky
 * update all copyright dates
 * update all .gold files with new test output
 * update ChangeLog in preparation for 1.0.0 release

at the end of their input.

Signed-off-by: Timothy J. Wood <tjw@omnigroup.com>

- Add new parser function, yajl_parse_complete().  This is defined
  to parse any buffered data assuming that there is no more input
  forthcoming and may issue callbacks based on that parse.

- Implement this as parsing a single space character.  This is
  sufficient to handle the problem case, a number at the end
  of the stream.

- Updated yagl_test.c to call this when the stream ends.

doesn't interpret this as an error it should handle, but something
that the higher level code should deal with in its representation.

Renamed the 'context' argument to yajl_lex_lex to 'offset' to be more in line with the other functions (to be used in a following commit).
Also renamed the local 'startCtx' to 'startOffset' to match.

These succeed right now, but that seems questionable since the the following
character is part of the same token.  A failure *would* be issued if the
parser knew that the stream was at an end, I think.  Still, the user of
the library might erroneously handle the null/true/false when yajl could
emit the error sooner.

This is useful for integrating with non-cmake build systems such as Xcode.