TimeStamp: from Scalar to unsigned long int?

I have been thinking in this apparently silly issue. Now we have a TimeStamp class based on a Scalar, thus, a double. Time stamps contain the seconds passed since 1970-01-01 midnight, now is something like 1535984912.645843518.

In ROS, TimeStamps are encoded by two unsigned int, one for seconds and the other for nanoseconds. I think maybe a unsigned long int encoding nanoseconds would be enough (until year 3330).

I think we should move to that implementation for two reasons:

• Compatibility: Now, when casting to Scalar a rounding to the double representation is performed changing a little. Then, comparing with ROS timestamps is not possible (getNanoSeconds() does not return the same value as initialized). This simple test does not work:

       unsigned long int sec = 5;
       unsigned long int nano = 1e5;
       wolf::TimeStamp t(sec,nano);

       ASSERT_EQ(t.getSeconds(),sec);
       ASSERT_EQ(t.getNanoSeconds(),nano);

• Equally check: Comparing two doubles with operator== is unrecommended (current implementation), however, we know that it may be useful (the same TimeStamp).
• Enough precision: We won't require more than nanoseconds precision, and Scalar hasn't a constant precision. Actually, since the seconds are a quite large number, I really don't know which precision do we have.

@jsola ,@asantamaria,@artivis are you ok with this change? Is there any reason to keep with the Scalar? Obviously, the API will remain unchanged.