Resolve vec4 : glm_vec4_distance() to satisfy compiling on armv7

[haxpor]

Fixed according to @recp 's suggestion of solution at #82 .
Test cases are added.

Tested on PC (Linux ubuntu 18.04 with SSE/SSE2/SSE3/SSE4 capable).
Tested on real Android device (Android 6.0, aarch64).

[haxpor]

note I noticed 1 failed test result in bezier (pulled from the tip).

[recp]

Hi @haxpor

Thanks for this PR.

I have added this to travis:

after_failure:
  - cat ./test-suite.log

we should see what is wrong in travis's logs now

[recp]

note I noticed 1 failed test result in bezier (pulled from the tip).

this is weird, it passes on my local macos and macos on the travis-ci. My Linux VMs are not working for now, after fixed that I'll test it on Linux

[haxpor]

I pulled down the tip from fix_82 branch again. I got the same result as before.

[       OK ] test_vec4
[ RUN      ] test_vec3
[       OK ] test_vec3
[ RUN      ] test_affine
[       OK ] test_affine
[ RUN      ] test_bezier
[  ERROR   ] --- glm_eq(smc, Bs_plain)
[   LINE   ] --- test/src/test_bezier.c:54: error: Failure!
[  FAILED  ] test_bezier
[==========] 12 test(s) run.
[  PASSED  ] 11 test(s).
[  FAILED  ] 1 test(s), listed below:
[  FAILED  ] test_bezier

 1 FAILED TEST(S)
FAIL test/tests (exit status: 1)

[coveralls]

Coverage decreased (-0.008%) to 11.081% when pulling e909c82 on haxpor:fix_82 into 73e6b65 on recp:master.

[recp]

SMC is matrix and vector multiplication. So there may be rounding errors, because SIMD registers (SSE/AVX....) do not have enough precision like single FPU... I was used this:

assert_true(glm_eq(smc, Bs_plain));
assert_true(glm_eq(Bs,  smc));

it must be:

test_assert_eqf(smc, Bs_plain);
test_assert_eqf(Bs,  smc);

Now all tests are passed.

[recp]

@haxpor thanks for your contribution!