Just thinking out loud here - I'm guessing that one lead of your thermocouple is 'grounded'. I wonder whether wiring it up to the AD595 in a way that keeps the differential inputs rather than grounding one lead of the thermocouple might improve things, by keeping any induced noise as a 'common mode' signal that would be rejected by the differential input.

Do you have access to an oscilloscope? I wonder if you could identify with greater confidence where the noise is entering the system by using the scope. Of course, sometimes the scope probe itself can introduce noise, and you may have trouble figuring out what 'ground' to connect to your scope probe.

Using a common ground for shielding is probably difficult to do in your case. The G540 is optoisolated, so you could probably establish a common ground on the output side of the G540 for the stepper shielding, any limit switches, and the like. But then what do you do with the metal frame of your machine, the stepper motor housings, and so forth? The Arduino's ground is likely to be common to the 120 VAC power ground, which would also be the ground for the PC and the input side of the G540. Whether or not the heater power supply and nozzle share that ground is anyone's guess.

But if you could characterize the noise better, perhaps you could knock it down with some kind of low-pass filtering, maybe a ferrite bead or two, for example. The thermocouple signal is bound to be a slowly-changing one, whereas any noise is probably going to be relatively high in frequency.