Fix and improve qint_from_int64_test

Fix and improve qint_from_int64_test

Use a long long integer constant to fix a compilation error (integer
constant is too large for 'long' type).

Use a better value for testing, as -1 makes the test pass even if
qi->value is of type uint32_t, float or double. This was suggested by
Reimar Döffinger <Reimar.Doeffinger@gmx.de>.

Also, make the test fail when qi->value is of type double or float by
casting qi->value to int64_t, to avoid value being promoted to the type
of qi->value.

Signed-off-by: Pierre Riteau <Pierre.Riteau@irisa.fr>
Acked-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>