mkimage.sh: loop-partition race — mkfs.vfat hits /dev/loopNp1 before the kernel publishes it

[VijitSingh97]

Seen cutting the RC20 release image on the release box while the KVM battery ran alongside: os/rauc/mkimage.sh failed at mkfs.vfat: unable to open /dev/loop0p1: No such device or address right after image allocation — the classic losetup -P race where the partition nodes aren't published yet, made likelier by loop-device churn from a concurrent battery run. A plain retry succeeded and verify-image passed 53/53, which is the fingerprint of a timing race, not a layout error.

Fix direction: after losetup -P, run udevadm settle and/or poll for the expected partition nodes with a bounded retry before the first mkfs. One guarded wait; the failure mode otherwise lands exactly when the box is busiest (release day).

🤖 Generated with Claude Code

[VijitSingh97]

Fix up in #827 — one guarded wait after losetup -P (udevadm settle + bounded poll for both partition nodes) before the first mkfs, mirrored in verify-image which opens the image the same way.