Handle NaN and infinity values

Right now, NaN/infinity values will produce "nan" and "inf", but
Python doesn't allow those in source code. This change will wrap
those values in float(''), which is allowed.

Tests for Python 2.7 and 3.8 have been added as well.

Fixes #136

bytecode.cpp

@@ -1,5 +1,6 @@
 #include "pyc_numeric.h"
 #include "bytecode.h"
+#include <cmath>
 
 #ifdef _MSC_VER
 #define snprintf _snprintf
@@ -268,7 +269,15 @@ void print_const(PycRef<PycObject> obj, PycModule* mod)
                                         obj.cast<PycComplex>()->imag());
         break;
     case PycObject::TYPE_BINARY_FLOAT:
-        fprintf(pyc_output, "%g", obj.cast<PycCFloat>()->value());
+        {
+            // Wrap any nan/inf values in float('').
+            double value = obj.cast<PycCFloat>()->value();
+            if (std::isnan(value) || std::isinf(value)) {
+                fprintf(pyc_output, "float('%g')", value);
+            } else {
+                fprintf(pyc_output, "%g", value);
+            }
+        }
         break;
     case PycObject::TYPE_BINARY_COMPLEX:
         fprintf(pyc_output, "(%g+%gj)", obj.cast<PycCComplex>()->value(),