"I see no purple light crashing out of you (so just walk on in!)" -- The Sisters of Mercy

 Notice anything different?

 It turned out as I suspected: there was one crucial place in the complex chain which, if I choked it, would block most of the noice from entering either the camera or the camera port in the Pi.  This is a lot easier to apply than adding components inside the TCS, but that still needs to happen since it is the source of the problem.

 But in the mean time, I got lucky: one common-mode choke (mystery ferrite, probably mix 31 based on how well it works) with two turns on the core.  I'd have done three but I couldn't get the connector through again.

 EDIT LATER: when I posted this, there was supposed to be a video clip showing the printer running via the R-pi cam, but without the purple flashes.  So here is that:

 This outcome confirms my suspicion that the EMI was A) conducted not radiated and B) common-mode in nature.  And those suspicions were just based on experience, I'm afraid.  I am startled to have found a critical spot on the first try.  Back when I was getting paid to do EMI suppression for "EMC Acceptance Testing" it was rarely this simple to solve a noise problem, I can assure you!  (fortunately, I had very smart and experienced help)

 At some point, I will install both common and differential mode filters, plus small capacitances (handwave let's say at the relay and at the outlet (forming a pi network I guess).

 One has to be careful specifying and selecting such caps to be used across the power line, especially if they're expected to survive surges and spikes...

 On a 120VAC circuit, a capacitor with a 300VAC rating would be a minimum for the supply voltage alone, and normal best practices suggest 600VAC rating.  From my own reading and experience, and given that there are no other transient voltage surge suppression devices in this box, these devices should be able to survive a 3kV spike if you're being lazy, and a 6kV spike if not.  Good thing these are very small value (.001uF/1000pF should do it, and the reactance is so low their current draw may be ignored) capacitors.  Some likely suspects are the self-healing metalized plastic units on DigiKey rated for 6kVDC, whose longest dimension is 1.2in - they should do nicely.  The ferrites will be somewhat larger.