Rebuild recovery-from-boot patch when calling add_img_to_target_files.

When using Verified Boot 2.0, releasetools specifies a salt value based
on build fingerprint, so that to give idempotent images.

However, the change that removed static `ro.build.fingerprint` [1] broke
the behavior, as common.LoadInfoDict still relies on fingerprints.
Without a fixed salt, the first call to make_recovery_patch.py and the
second one (which writes IMAGES/{boot,recovery}.img) will see different
images, which leads to install-recovery.sh failure.

Note that currently there's a dependency that requires getting bootable
images through two separate calls. make_recovery_patch.py has to happen
first to get (placeholder) files in the system image. We then generate
canned fs_config files, and finally use add_img_to_target_files.py to
write the images.

This CL adds a quick workaround to force rebuilding the
recovery-from-boot patch while calling add_img_to_target_files.py.

[1] https://android-review.googlesource.com/c/platform/build/+/892933

Bug: 134123803
Bug: 134525174
Test: TreeHugger
Test: Build a non-A/B target that uses AVB. Run validate_target_files.py
      on the generated target_files.zip.
Change-Id: I5859e30be63bfd54398cf41fd2d907f15285f560