In particular this allows us to get rid of our custom implementation
for the opacity transition (which did not always work properly).

This avoids caching "wrong" values when setting defaults in
Preferences::_loadDefaults().

(we did this since 7ca9dc4c53873955a31c1b71559da9f7b29b4a2e for
 OS-specific defaults)

Fixes https://gitlab.com/inkscape/inkscape/issues/99

Fixes https://gitlab.com/inkscape/inbox/issues/850

Those should be used for RGBA color values, which would overflow
the signed type otherwise.

Replacement will be seamless and fully backwards-compatible, as we
can read the overflowed (i.e. negative) signed integers we used to
save and they will be properly unwrapped.

Also reading the unsigned ints as signed ints in older versions of
Inkscape will still work, as we always depended on the undefined
behavior of atoi in case of out of range numbers (which luckily is
means overflowing in gcc).

As we handle unsigned integers the same way as signed integers,
large unsigned values (notably RGBA values) overflow the signed type.

The issue became apparent when we switched from atoi to strtol in
  0e92e8e519b33b35ab63600c31feb7c639e347a6

While the former has undefined behavior for out of range numbers
it luckily seemed to "just" overflow as expected on supported OSs.

The latter however returns LONG_MAX (which for a 32-bit value is
0x7fffffff and turned out to show turquoise in unexpected places,
e.g. the default path highlighting color in node tool).

Avoid this by catching the error and try to parse as unsigned type
in this case (which will still overflow, but seems to have worked
as expected so far).

Fixes GitLab issue 128.