Avoid inaccurate floating-point computations

...otherwise at least my --disable-dbgutil --disable-debug Linux x86_64 build
failed the CppunitTest_sax_cpputest with 8999999 vs. 9000000 nanoseconds.

Change-Id: I05e0febf413f9f9e01227a0cc4e0f46a5243fe61

sax/source/tools/converter.cxx

@@ -1122,9 +1122,13 @@ bool Converter::convertDuration(util::Duration& rDuration,
                     {
                         if (-1 != nTemp)
                         {
-                            const sal_Int32 nDigits = std::min<sal_Int32>(nPos - nStart, 9);
-                            OSL_ENSURE(nDigits > 0, "bad code monkey: negative digits");
-                            nNanoSeconds=static_cast<double>(nTemp)*(1000000000.0/pow(10.0,nDigits));
+                            nNanoSeconds = nTemp;
+                            sal_Int32 nDigits = nPos - nStart;
+                            assert(nDigits >= 0 && nDigits <= 9);
+                            for (; nDigits < 9; ++nDigits)
+                            {
+                                nNanoSeconds *= 10;
+                            }
                             nTemp=-1;
                             if ('S' == string[nPos])
                             {