fix: propagate boot image flash failures

[rawer886]

Summary

• propagate boot image unpack, patch, repack, write, sync, and verification failures
• preserve flash_image exit codes in direct install, installer, and unpatch flows
• verify uncompressed images after writing before reporting success
• prevent unpatch cleanup commands from hiding boot image restore failures
• distinguish a successful boot image restore from a later APatch file cleanup failure

Background

Direct Install could report Successfully Flashed even when dd failed to write the boot partition. flash_image discarded dd stderr and returned success after sync. Several callers also read $? after another command had already replaced the original result.

The issue was reproduced with a boot partition write returning I/O error and SCSI Data Protect; the boot partition hash remained unchanged, while the same patched image flashed and booted successfully through fastboot.

Changes

• return distinct failures for insufficient partition size, read-only targets, write/sync failures, and post-write verification failures
• retain the original exit status immediately after unpack, patch, unpatch, repack, and flash commands
• verify the written prefix of uncompressed images with the packaged BusyBox cmp -n implementation
• stop the unpatch flow before cleanup when boot image restoration fails
• run cleanup only after the boot image is restored, while still attempting both cleanup operations and reporting any failure
• keep the reboot-required state when the boot image was restored even if subsequent cleanup fails

Validation

• sh -n passed for all changed scripts
• git diff --check passed
• :app:compileDebugKotlin passed
• :app:assembleDebug passed
• :app:testDebugUnitTest passed (NO-SOURCE)
• packaged BusyBox 1.36.1.1-topjohnwu was verified on the affected beryl device to support cmp -s -n; equal and different prefixes returned 0 and 1 respectively
• device-side failure injection under the packaged BusyBox ash verified that unpack, patch, unpatch, and repack failures preserve their original non-zero exit codes
• device-side cleanup injection verified that both cleanup operations are attempted and the combined command remains failed when either operation fails

Fixes #1585