Make sure that range limit of last thread never exceeds data size

Support compiling with clang on windows

THUDNERX2T99: Fix clang compilation

CMake Improvements

Travis: Rewrite config, build and test also on Alpine Linux (musl libc)

See: https://github.com/xianyi/OpenBLAS/pull/1255#issuecomment-318692183

See: https://github.com/xianyi/OpenBLAS/pull/1255#issuecomment-318628666

Alpine jobs needs sudo (for chroot), so they run on VMs infrastructure.
That's why they are much slower than other jobs.

Using APT addon has nice side-effect - you don't need sudo anymore, so
it can run on Travis containers-based infrastructure that is much faster
than their VMs infrastructure (used when sudo is needed).

You've been still running on Ubuntu Presty builders, but new default is
Trusty. Thus I've explicitly set `dist: presty` to let it stay on
Presty, to not change build environment by this commit.

Revert "Fix calculated range limit exceeding actual data size for last thread"

Rework __GLIBC_PREREQ checks to avoid breaking non-glibc builds

Fix calculated range limit exceeding actual data size for last thread

Honor cgroup/cpuset limits when enumerating cpus

Use cpuid 4 with subleafs to query L1 cache size on Intel processors

0.2.20 version