Your schematic has problems:
- Self heating of the thermistor may cause a few degrees of error. (Probably a small niggle, but it causes VCC sensitivity, and is an easily solved design issue.)
- The op-amp is fixed gain, so the accuracy won't be good anyway.
- Putting caps on op-amp inputs is a bad idea.
The darlington has the same gain as the MOSFET -- the source/emitter degeneration resistor dominates. Either is fine. I didn't have a PMOS handy for that project, and BJTs are cheaper anyway.
SMTs for an oven are kind of dubious, just because you don't have any good way to spread the heat out. In my project, the regulator and transistor were soldered to a sheet of, I think, 8oz copper, which wrapped around the PCB.
D/D2PAKs could be soldered to a heat slug very nicely, but you need to arrange that somehow. (You can even get PCBs with heavy copper, and slugs, and stuff in them, but... yeah... custom $$$...)
Anyway, FETs comparable to IRF510 (which itself is stupid cheap!) are available in anything you can think of: SO-8, SOT-223, SOT-89, DPAK... Likewise, TIP32C --> MJD32C (DPAK).
LM358 will be an okay replacement for TLV2372. It obviously works fine under the 5V conditions, and is okay for 12V, but not if it's dirty 12V. Nice thing about the bipolar output: it biases the op-amp's output stage, which eliminates LM358 crossover distortion. That's something you might miss with a MOSFET output.
Tim