Floating Point Variables Are Ignored

Version 2.4.2846 (Build ID df7c027e)

Describe the bug
Floating point variables seem to be ignored in the Binary Ninja Intermediate Language.

To Reproduce

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

float test3() {
    srand(time(NULL));
    float f_x = 41.89465555f;
    float f_y = (float)(rand()) / 100000.0f;
    float sum = f_x + f_y;
    return sum;
}

int main() {
    float x = test3();
}

1. compile with gcc -g0 -std=c99 -fno-stack-protector -m64 -O0
2. open compiled binary
3. navigate to test3 symbol
4. show Medium Level IL

Result:

       00001279  test3:
   0 @ 0000128a  rax = time(nullptr)
   1 @ 0000128f  rdi = rax
   2 @ 00001291  srand(x: rdi)
   3 @ 00001296  zmm0 = 41,8946571f
   4 @ 0000129e  var_c = zmm0
   5 @ 000012a3  rax_1 = rand()
   6 @ 000012a8  zmm0 = 0f
   7 @ 000012ac  zmm0 = float.s(rax_1.eax)
   8 @ 000012b0  zmm1 = 100000f
   9 @ 000012b8  zmm0 = zmm0 / zmm1
  10 @ 000012bc  var_10 = zmm0
  11 @ 000012c1  zmm0 = var_c
  12 @ 000012c6  zmm0 = zmm0 + fconvert.t(var_10)
  13 @ 000012cb  var_14 = zmm0
  14 @ 000012d6  return rax_1

Expected behavior
The return value of the above example should be var_14 instead of rax_1.

Version and Platform

• Binary Ninja: Version 2.4.2846 (Build ID df7c027e) / Version 2.4.2917-dev (Build ID d828c284)
• OS: Ubuntu Linux
• Version: 20.04.2 LTS

Additional context

• binary_ninja_bug_report.zip several examples (including above mentioned one) that are related to this issue (source and compiled binary)
• Could be related to #2566

Start with the C reproducer and compiled examples in binary_ninja_bug_report.zip, then inspect the test3 symbol in Medium Level IL. Compare the generated return value with var_14 and investigate why rax_1 is returned instead. Done means floating-point test3 returns the computed var_14 value in the IL.