Note that altlvds_rx with a deserialization of /2 will be built with generic logic, i.e. an altddr_input block to capture the data, and a PLL that drives a global clock tree. The timing is not very good with this layout(although I believe a large portion is due to overly-pessimistic timing models for skew analysis). If you go to a higher serialization rate, such as /4, the timing is excellent. I think you get a sampling window of 300ps, and this is because all the dedicated LVDS hardware is used(a very low skew, dedicated clock tree, a register that samples at the higher rate so there's no rise/fall variation). I've seen on more than one occasion where a user that really only needs /2 deserialization instead build an altlvds block with /4 deserialization.
As to your question, the data is serial, so just looking at it by itself you could never determine the order. I believe what you're asking is it's relation to the external clock. The MSB output is what is captured by the rising edge, so if you'r sending clock/data center-aligned, it's the one that lines up with the rising edge. If sending edge-aligned I believe it's the one coming after the rising edge, since the clock will be shifted 90 degrees in time. (If you're simulating, you'll know in a second, and even in hardware you'll probably be able to quickly determine if it's right or not, and its easy to switch.)