Sorry, you're quite right. The issue is that there are two conflicting considerations.
(1) Higher initial selectivity on the index. This is quite valid, but it's a general consideration, a "default" rule if you will.
(2) Avoid fragmentation of the clus index. This is valid too, and in this case I believe it overrides (1), since both columns will (almost) always be specified anyway and because c2 appears to be relatively random.
Index tuning often involves a number of judgment calls. I'd accept another DBA preferring c2 first to c1. Honestly you'll probably be OK with either. As long as you do not fall back on an identity column instead, that would be a big mistake here.