Using optimized byte swaps reduced portability for no real benefit,
since they are in parts of the code that represent a tiny fraction
of the execution time.  So a simple definition of a byte swap is
now used.

SunOS 4.1 claims that it is __STDC__, but it does not have strerror
in string.h.  Instead of using __STDC__, this puts a direct test
for strerror in configure, and uses that information in gzguts.h.

SunOS 4.1 doesn't have memmove(), and there may be others.  memcpy()
should not be used for overlapping copies, so here a simple copy is
implemented that works for the particular direction of the overlap,
which is where the destination precedes the source.

Apple removed support for gcov in the default gcc compiler chain,
when they moved to llvm.  This can be circumvented in XCode 4.2 by
using the gcc chain with gcc-4.2.  This patch allows setting
GCC_CLASSIC to the name of a real gcc executable (e.g. "gcc-4.2")
to allow coverage testing.

crc32.c was #including limits.h in order to find a four-byte integer
type.  It was doing this even if Z_SOLO were defined, violating the
intent of Z_SOLO, which is to include no library headers and require
no library functions.  Now crc32.c obeys the intent of Z_SOLO, but
with the downside that crc32() will be slower than when not compiled
with Z_SOLO.  This can be remedied manually by typedefing u4 to a
known four-byte unsigned integer type, and #defining BYFOUR in
crc32.c.

gzflags() was put in gzwrite.c in order to be compiled exactly the
same as gzprintf(), so that it was guaranteed to return the correct
information.  However that causes a static linkage to zlib to bring
in many routines that are often not used.  All that is required to
duplicate the compilation environment of gzprintf() is to include
gzguts.h.  So that is now done in zutil.c to assure that the correct
flags are returned.