I think the way a DM decides this (insignificant) conundrum has a lot to do with his agenda.
If you make a blanket statement about Stealth and Blindsight, either you make it public to the player, and you are in the realm of "WotC D&D tournament" which looks a lot like MtG, or you keep it on your side of the screen, and you are in the realm of "Gygaxian skilled play", where knowledge has to be earned through trial and error.
If you intend (as I am feeling inclined to at this particular time) to resolve the conundrum at the table, you are closer to Dungeon World agenda, where you play to know how the Dragon reacts to the thief crouching behind the boulder. The decision should involve dice rolling, the Stealth and Detection scores, blabla... I believe this is the (not hidden) intent of the designers, and given the conceit and actual numbers involved in bounded accuracy, without the DM taking sides with ad/disad, the odds of rolling high or low are not that far from DW's.
I would take the comparison further : the Legendary framework enables the DM to mitigate successes and failures. Instead of denying the Rogue one of its favorite tactics, or making dragons flying bags of XP by sacrificing them on the altar of Ambushing, it becomes easy to counter the Rogues attempt at stealth without shutting them down. For instance, maybe the Dragon knows the Rogue is somewhere, and spends a LA to Detect it (maybe with Advantage). The rogue won't sneak attack, but the party avoids a tail slap or wing buffet. Or maybe the dragon focus on the boulder/the rogue, buying some time for the party. Or, maybe, the Dragon is so focused on destroying the remainder of the PCs that it forgets about the rogue and let her sneak attack the next round... Note that if you want to promote this gameplay without demoting the others, offering vague rules is a clever plan.