Hi,
It seems likely that this is a nuance of the ever increasing distribution of each col2 value. What is does kind of show is that when you build an index it doesn't build an index and then run update stats it must builds the histogram as it goes. The histogram after create index looks like it has been built making the assumption that the density of values in the initial part of the table are representative of the whole table, but this is not the case. As such it might be that the stats are built based on how the sort is operating on the data, not entirely sure.
I know there are multiple phases involved when ASE builds a histogram, the last of which does some sort of reduction in an attempt to normalize the histogram.
When you update using the index, it uses the leaf level of the index to build the histogram which holds far more information pertaining to the upper and lower bounds and therefore can build a histogram from a 'better' starting point. I say 'better', because if anything the histogram after index creation has too many uncessary steps.
I think this would be worth reporting to SAP based on you reproduction to get the offical take on it, there is a chance that maybe a phase of histogram processing isn't being completed on index creation or something else they need to improve upon.
Cheers,
Simon