On the radiator:
I think I'd be scouring junkyards to find a suitable core. That's actually a pretty large piece--bigger than what's in my Valiant, and that car stays cool. Can't clear the cap? Zip it off flush and solder/weld on a patch (or have a shop do it). Then use either a later-model overflow/recovery tank with a pressure cap (these are getting pretty common) or use an inline pressure cap adapter like
this one. Or, if you want to roll the dice a bit,
eBay is full of 'em in various hose diameters (even stepped) and styles.
Exhaust: The muffler is your restriction. There's no need to make the pipe any larger than the muffler's smallest nipple. That being said, if you're doing crush bends rather than mandrel, I'd go up at least one pipe size. The pipe's cross-sectional area is much smaller in the bend area on non-mandrel bends. I can't help you with the cutouts; it's not something for which I've shopped, nor have I ever sold any. For the Valiant I made dump pipes terminated with V-band caps. Manual, but fast to remove. Good electric ones are expensive, cheap non-electric ones are cheap (and usually made of cast iron, for some reason).
LED bulbs: I had those exact "Cougar" LEDs in my Dakota. They were about $9/pair on Amazon at the time. I also had the 1156 version for the backup lights, which were ridiculously bright--think about how small the backup light is on an '80s Dodge pickup or Dakota through '96 (same taillight). They were nearly rear-facing fog lamps. I had no issue with quality; they worked well. That Cougar site is stroking people.
Other than known vendors of ITB or stack injection, I wouldn't know where to point you for trumpets. I'll just remind you that the length directly and noticeably affects your powerband. The longer the pipe, the lower the RPM range. While I'm on that subject, and I'm sure we've discussed it previously, there's no truly good way to do speed-density tuning on ITBs. You're essentially limited to Alpha/N tuning (RPM v throttle position), unless you want to drill a vacuum port on each runner below the throttle blade, run a hose to a
large common vacuum chamber, and read MAP off that. Even that solution is a bit sketchy; ITBs are notoriously "noisy" due to the constant pulsing between vacuum, reversion and ambient air pressure as the intake valve opens and closes. Alpha (Alfa?)/N is a lot of work to properly implement, but the throttle response of ITBs can't be beat. There is literally no adaptive ("self-learning"--
HA!) algorithm that will work decently with stack injection.