As someone who has spent an unfortunate amount of time working on shipping label software for ecommerce companies, my advice is to stay the hell away from ZPL. It's the layout language equivalent of C -- too many low-level footguns. Just send 4x6 PDF's to your thermal printer and be done with it. There are so many more tools, of much better quality, available for composing PDF's than you'll ever find for ZPL or other vendor-specific layout languages.

Sure, if you know you're always going to be using a Zebra printer, and you control the whole pipe from software to the physical label, then you're going to have a lot more explicit control if you use ZPL over a PDF. But the second any of those presuppositions change, it's back to the drawing board. (Actually, even the Zebra assumption isn't sound, because they themselves are not always 100% reliable in following their own specifications. And the specifications... boy howdy.)

The old argument in favor of ZPL is that you can be much more precise with the instructions you're sending to your thermal, resulting in higher print quality, and therefore more reliable barcode scans and OCR. But the world doesn't stand still, and shipping carriers have been steadily improving their scanning capabilities over the last 20 years, even while thermal printer capabilities have remained largely stagnant. After all, USPS has to also optimize for people who are printing labels on shitty desktop inkets. The increase in barcode quality you'd get out of using ZPL is overkill.

TLDR: Save yourself a lot of heartache and just sling PDF's. Postscript/Ghostscript is so much easier to cajole into cooperation.