test: make ValueTest/objects float check robust against x87 excess precision (#1700)

ValueTest/objects stores the float literal 0.12345f into a Json::Value
(widened to the stored double) and then asserts equality against the
original 0.12345f. On 32-bit x86 targets where GCC defaults to x87 math
(-mfpmath=387, 80-bit excess precision) and the baseline lacks SSE2, the
round-trip yields the exact double 0.12345 rather than the float-widened
0.12345000356435776, so the exact double comparison fails even though the
library itself is correct.

Compare the stored value narrowed back to float via asFloat(): both the
expected literal and asFloat() collapse to the same float value on every
architecture, keeping the numeric round-trip check while making it
precision-robust.

Co-authored-by: Jordan Bayles <bayles.jordan@gmail.com>

Author: plusky

src/test_lib_json/main.cpp

@@ -352,7 +352,7 @@ JSONTEST_FIXTURE_LOCAL(ValueTest, objects) {
 
   const Json::Value* numericFound = object2_.findNumeric("numeric");
   JSONTEST_ASSERT(numericFound != nullptr);
-  JSONTEST_ASSERT_EQUAL(0.12345f, *numericFound);
+  JSONTEST_ASSERT_EQUAL(0.12345f, numericFound->asFloat());
   JSONTEST_ASSERT(object3_.findNumeric("numeric") == nullptr);
 
   const Json::Value* stringFound = object2_.findString("string");