tvdiff_secs(): sub-zero time difference adjustment

Skip a floating point addition operation when integral part of time difference
is zero. This avoids potential floating point addition rounding problems while
preserving decimal part value.