Installer fixes: dual-boot free-space + full-disk (#3 #115 #127 #130 #137)

[nightdevil00]

Installer fixes: dual-boot free-space + full-disk installers

Closes #3, #115, #127, #130, #137.

Seven installer defects across both install workflows, each fixed so a failure
degrades to the next mechanism instead of aborting the install, and none
changes the standard (non-broken) path.

Dual-boot free-space ([NTFS][free space])

• Different-disk messaging. The free-space flow already creates its own FAT32
  ESP + root in the selected gap (it never touches Windows' ESP/ext4). The disk
  configurator now says so on-screen, removing the mis-targeting that aborted
  installs on a second disk.
• Stale btrfs/LUKS signature on reinstall (Fix omarchy-iso-boot #3). A single one-shot
  wipefs -af raced udev and aborted the reinstall. New retried,
  error-tolerant clear_stale_signatures() (up to 3×, returns 0 on failure)
  replaces it in run_partition_execute.
• Full NVRAM blocks the Limine entry (Install fails on old UEFI when efibootmgr refuses to create a boot entry, and the reason is discarded #127). efibootmgr --create failure
  used to discard the real error and either roll back or abort. Now the reason is
  captured, dangling entries are purged (Windows/live untouched), the create is
  retried once, and on final failure Omarchy installs the
  \EFI\BOOT\BOOTX64.EFI fallback, enables Limine's fallback, keeps the pacman
  hook in sync, and validate_boot accepts the missing entry only when the
  fallback exists.

Full-disk

• Stale LUKS header aborts archinstall's wipe (Full Install aborts when the target disk already holds a LUKS partition #137).
  omarchy-iso-cleanup-disk now wipes partition + disk signatures itself
  (retried, best-effort) after releasing holders, so archinstall's fragile
  per-partition wipe is replaced by a clean create.
• cryptsetup close "Device root is still in use" race (Omarchy 4.0.1 install intermittently fails with Device root is still in use #130). The adapter's
  existing udev-race retry now also recognizes the busy-device close message,
  force-closes stray crypt mappings, and retries — scoped to that exact message
  so genuine failures are still not looped.
• archinstall scans a LUKS partition as btrfs and crashes (Installer crashes with DiskError when mounting LUKS partition as Btrfs during device scan (NVMe + LUKS + Btrfs) #115).
  get_btrfs_info() is wrapped (try/finally-restored) to skip crypto_LUKS
  devices during the partitioning operation, suppressing the bogus
  "wrong fs type, bad superblock" mount. If the archinstall symbol can't be
  found the patch self-disables and behaves as before.

Tests

• test/unit/test_nvram_fallback.py — Install fails on old UEFI when efibootmgr refuses to create a boot entry, and the reason is discarded #127 (uuid parsing, dangling-vs-live
  reclaim, _NvramWriteError with "No space left on device").
• test/unit/cleanup-disk-wipe-test.sh — Full Install aborts when the target disk already holds a LUKS partition #137 (stale signature wiped on image,
  clean no-op, busy device tolerated by the call site).

All Python and shell unit tests pass.

Notes for reviewers

• The Installer crashes with DiskError when mounting LUKS partition as Btrfs during device scan (NVMe + LUKS + Btrfs) #115 monkeypatch is deliberately version-tolerant and scope-limited to the
  partitioning operation (restored in finally). Open to alternatives if you'd
  rather pin a specific archinstall version.
• Full Install aborts when the target disk already holds a LUKS partition #137/Fix omarchy-iso-boot #3 wiping is deliberately best-effort: a busy device must never trade
  archinstall's old abort for a new one.