If we're talking theory, the pixel size doesn't matter - it's the amount of information encoded in the QR code that matters (ie. excluding ECC bits!). In turn, that means it is theoretically possible to decode a QR code where each 'pixel' of the code is smaller than the pixel of the camera sensor observing the code.

While it is theoretically possible, nobody has yet done so...

Why? Is this something that's a side effect of some irrefutable physics law or a limitation of the current technology?

Both. You can't have a measurable feature smaller than the Planck length, but that's not relevant to QR decoding in any practical sense.

As for the practical limits, try using your phone to take a picture of something very far away, or very tiny, and see what happens.