I've been seeing inconsistent results for

RepositoryTests.test_password_decryption_failed, which tests that we

don't crash when the SECRET_KEY has been mistakenly changed and stored

passwords cannot be decrypted. While this works most of the time by

properly catching the decode error, occasionally the random IV used

during AES-CFB causes the decoded result to be valid(ish) unicode. These

almost always result in unprintable characters, however.
This change adds a step to decrypt_password to check if the result is

printable, and if not, raise a ValueError. This should make it so this

test (and handling of the situation in production) is much more

reliable.

Ran unit tests.