math: simplify type conversion with casts and limits checks

This PR aims to address some issues presented in Fluent Bit
repo:

 - fluent/fluent-bit#8083
 - fluent/fluent-bit#8762

Signed-off-by: Eduardo Silva <[email protected]>

include/cmetrics/cmt_math.h

@@ -21,12 +21,7 @@
 #define CMT_MATH_H
 
 #include <inttypes.h>
-#include <string.h>
-
-union val_union {
-    uint64_t u;
-    double d;
-};
+#include <math.h>
 
 /*
  * This is not rocket-science and to make things easier we assume that operating on
@@ -35,18 +30,24 @@ union val_union {
 
 static inline uint64_t cmt_math_d64_to_uint64(double val)
 {
-    union val_union u;
+    if (isnan(val)) {
+        return 0;
+    }
+
+    if (isinf(val)) {
+        return UINT64_MAX;
+    }
 
-    u.d = val;
-    return u.u;
+    if (val > (double) UINT64_MAX) {
+        return UINT64_MAX;
+    }
+
+    return (uint64_t) val;
 }
 
 static inline double cmt_math_uint64_to_d64(uint64_t val)
 {
-    union val_union u;
-
-    u.u = val;
-    return u.d;
+    return (double) val;
 }
 
 #endif