Float and double conversions when logging

[anjohnson]

Another bug we've found at APS which I was planning to fix when I have time: The code here for both JSON and old-style logs uses the simple printf() format string "%g" to convert both double and float values. This loses precision for doubles at least, which matters for some applications.

The yajl JSON generator uses "%.17g" for converting doubles and appends ".0" if the resulting string only contains digits and/or a minus sign, which may be necessary for strict JSON. We don't care about the old-style logging format, but we do want the JSON output fixed.

This is the code used by yajl_gen:

sprintf(i, "%.17g", number);
if (strspn(i, "0123456789-") == strlen(i)) {
    strcat(i, ".0");
}

The caPutLog code has already handled ±Infinity and NaN values so they never get to the fieldVal2Str() routine.