Collector = substrate isn't always used, because that can be very tricky for loose charge carriers diffusing into the control circuitry.
Substrate is usually the lowest or highest voltage node, depending on if it's N or P type.
Wells are diffused into the substrate, then transistors into that. Presumably, the LM317 connects output to substrate for convenience, and it would have to be P substrate for an N-well (output transistor collector), with P and N diffusion/epitaxy on top of that (making the rest of the transistor). Which, as it happens, means there's a huge friggin' reverse-protection diode there: the N-well (input) to P-substrate (output). It might not be desirable to use, though (poor current distribution, no guard rings..?).
This makes PNPs very difficult, because they either have to be substrate = collector (pretty useless), or they have to be diffused on a P-well in an N-well, which is a triple liability: the doping on the emitter has to be terribly high, the collector voltage rating will only be the NPN emitter rating, i.e., ~7V, and the PNPN layering of the wells allows catastrophic SCR latchup to occur (the PNP on top may not be usable at all because of charge diffusion away from it). This is why they invented "lateral PNP": shit hFE, but symmetrical C/E ratings (30V!), good enough for diff pairs, current mirrors and such.
I suppose LM337 could be a layer-for-layer swap (substitute P for N), though I doubt you can do that strictly so easily, because the dopants have different diffusion rates so you need to tune all the process steps again... That's probably part of the reason why they used a different design (it's actually an LDO, just not as useful because it's negative!).
LM1117 is perhaps where things get interesting: it's more like a positive '337. I forget if that was one fuckoff-massive lateral PNP, or a single NPN follower (not Darlington) and a slightly-less-massive lateral PNP to drive it. I think it's still made on the classic process, not a complementary one?
Oh, so, anyway, note that tab = substrate isn't necessarily any internal connection, because nodes can be isolated through various means, allowing voltages above and below substrate to be used inside. That's why the LM317 substrate is neither the collector (N substrate for a power device would actually be rather inconvenient, for the above reasons) nor the emitter (the diagram shows emitter resistance, for current sharing -- it's actually dozens of emitter 'fingers', wired in parallel with a little added resistance), but "ground" of a sort.
Tim