kickstart: Fix multi-ESP install and GRUB boot on RAID 1

Three issues prevented a successful Anaconda-based bootc install onto a
RAID 1 array with per-disk ESP partitions:

1. Bind mount failure during ostreecontainer payload install:
   Anaconda mounts the root FS at /mnt/sysimage and then tries to
   bind-mount /mnt/sysimage/boot/efi2 and /boot/efi3 into /mnt/sysroot.
   However, it does not create the mount point directories for the
   secondary ESPs before the bind mount, causing:
     mount: /mnt/sysroot/boot/efi2: special device
            /mnt/sysimage/boot/efi2 does not exist.
   Fix: add a %pre-install script that creates the directories after
   the root FS is mounted but before the payload runs.

2. Empty secondary ESPs after installation:
   Anaconda only installs the bootloader (shim, GRUB, grub.cfg) to the
   primary ESP at /boot/efi. The secondary ESPs (efi2, efi3) are
   formatted and mounted but receive no bootloader files. Additionally,
   the 'noauto' fsoption causes Anaconda to unmount them before %post.
   Fix: add a %post --nochroot script that explicitly mounts each
   secondary ESP by device path and copies the full EFI directory tree
   from the primary ESP.

3. GRUB cannot find the kernel on the RAID root filesystem:
   The GRUB EFI binary shipped with CentOS Stream 10 does not include
   the mdraid1x module. GRUB's grub.cfg uses 'search --fs-uuid' to
   locate the root filesystem and load the kernel from the ostree
   deployment path, but it cannot access the md array.
   Fix: add a separate /boot partition (ext4, 1G) on the primary disk.
   GRUB can read ext4 natively without mdraid support. The kernel and
   initramfs live on /boot, and the initramfs handles RAID assembly
   for the root filesystem during boot.

Note: /boot is only on vda (not mirrored), so only disk 1 can boot
independently. Full boot redundancy from any disk would require either
whole-disk RAID (where GRUB reads from the raw mirror) or a GRUB build
with mdraid1x support. This is a known limitation tracked in:
  - bootc-dev/bootc#1911
  - coreos/bootupd#1077

Author: jmarrero

kickstart.cfg.in

@@ -12,6 +12,8 @@ part /boot/efi  --fstype=efi --size=600 --ondisk=vda
 part /boot/efi2 --fstype=efi --size=600 --ondisk=vdb --fsoptions="noauto"
 part /boot/efi3 --fstype=efi --size=600 --ondisk=vdc --fsoptions="noauto"
 
+part /boot --fstype=ext4 --size=1024 --ondisk=vda
+
 part raid.01 --size=1 --grow --ondisk=vda
 part raid.02 --size=1 --grow --ondisk=vdb
 part raid.03 --size=1 --grow --ondisk=vdc
@@ -21,3 +23,38 @@ raid / --device=md0 --fstype=xfs --level=1 raid.01 raid.02 raid.03
 ostreecontainer --no-signature-verification --url=@@IMAGE_URL@@
 
 poweroff
+
+%pre-install
+# Create mount point directories for the extra ESPs.
+# Anaconda mounts the root FS at /mnt/sysimage before payload install,
+# but may not create subdirectory mount points for non-primary ESPs.
+mkdir -p /mnt/sysimage/boot/efi2
+mkdir -p /mnt/sysimage/boot/efi3
+%end
+
+%post --nochroot --log=/mnt/sysimage/var/log/ks-post-esp-copy.log
+# Replicate the bootloader files from the primary ESP to the extra ESPs.
+# Anaconda only installs the bootloader on the primary /boot/efi; the
+# secondary ESPs need identical content for redundancy.
+#
+# The extra ESPs have fsoptions="noauto" so Anaconda unmounts them before
+# %post runs. We mount them explicitly by device path here.
+set -euxo pipefail
+
+PRIMARY="/mnt/sysimage/boot/efi"
+
+# vdb1 = efi2, vdc1 = efi3
+EXTRA_DEVS="/dev/vdb1 /dev/vdc1"
+
+for dev in $EXTRA_DEVS; do
+    mnt=$(mktemp -d /tmp/esp-copy.XXXXXX)
+    echo "==> Mounting ${dev} at ${mnt}"
+    mount "$dev" "$mnt"
+    echo "==> Copying EFI content from ${PRIMARY} to ${mnt}"
+    cp -a "${PRIMARY}/." "${mnt}/"
+    echo "==> Contents of ${dev}:"
+    ls -R "$mnt"
+    umount "$mnt"
+    rmdir "$mnt"
+done
+%end