Some CPUs advertise RDRAND in CPUID, but return
0xFFFFFFFF unconditionally.  To avoid locking up
later, test RDRAND during initialization, and if
it returns 0xFFFFFFFF, mark it as nonexistent.

Also fix a possible segmentation fault in CPUID check.

This commit is a squashed and slightly modified backport
of the following commits on the master branch:

  * 0e5bbcaa16
  * 4d36b0287d
  * 8086314026

2.replace json_object_get/put API with json_object_retain/release, as they operate the reference counter, and confused with array_list_get/put_idx.
3.replace array_list_get/put_idx API with array_list_get/insert to make them more clear to use.

If we want to override `strerror()` in libjson-c
to make tests consistent across platforms, we
need to do it build-wide as configure/build
option.

Apple linkers make it really hard to override functions
at link-time, and this seems to be locked down on travis-ci.org
[ for security reasons I assume ].
While I got it to work locally, it did not work
when running on travis.

Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>

Remove trailing whitespaces

Fix build using MinGW.

MinGW requires wincrypt.h.
GCC does not support #pragma comment, which trips Werror.

* CVE-2013-6371: hash collision denial of service
* CVE-2013-6370: buffer overflow if size_t is larger than int