Both and neither. The fluff should absolutely influence the mechanics, while the mechanical framework will also necessarily constrain the fluff.
Ideally, the game should be constructed with mechanics that are as flexible as possible, allowing the designers to then describe whatever fluff they want, and be able to convert that to a mechanical representation. (This may cause some problems if, for example, they want to write an adventure featuring 1st level PCs vs a Great Wyrm...)
In reality, of course, there are limits to how far these things can stretch, and there needs to be some sort of compromise.
Ultimately, I would argue that the over-riding design goal for a new edition of D&D should be "feels like D&D". What I don't necessarily agree with is that that automatically requires the game retain, for example, Vancian magic, especially as the only option for magic. The 3e Sorcerer/Wizard split, for example, was at least an attempt to be a bit more broad in what the system can handle; it was a noble attempt, even if it ultimately fell short.
After the core system has been released, I would also argue that the mechanics should then be considered primary over the fluff; if there's something that the mechanics simply can't stretch to accomodate, it should probably be removed from the fluff of the game. That said, with the ability to add new classes, spells, monsters, rituals, and even whole new power sources (perhaps with their own mechanics), there's a whole lot of flexibility available there!