Would extending optimisation data field to other markets be a proper way to deal with this problem?
That's exactly what I'd do. I use the 10/21 combination because I have found it to be effective across decades of data from other markets. In other words, I can trust this combination to help me know when to trade most liquid asset classes. It let's me profit from the average trend.
So wouldn't it be logical to optimise the system with a stop loss already included rather than optimising the system without a stop loss first and then add a stop loss into a system which is already optimised without a stop loss, thus trading the system not the way it was tested?
In my opinion, there isn't a "correct" way to do it. I have heard reasoned arguments for both ways (indicators first followed by stop-loss or all at once). In my opinion, all that matters at the end of the day is that you have a system that catches the average trend and a fail-safe exit to protect you from wipe-out.