Fix bug where attr_writer was overriding the key_root= method.

[jrafanie]

Extracted from #4082

@kbrock please review...@Fryguy and I fixed an issue where the key_root= class method was being

✌️ whacked ✌️

by the attr_writer, causing the v*_key caches to not be reset via key_root=, particularly during tests.

I ran into this "bug" locally for seemingly unrelated tests and couldn't track it down previously:
bin/rspec gems/pending/spec/util/miq-password_spec.rb spec/models/miq_server/rhn_mirror_spec.rb --seed 46323

It's weird to run gems/pending tests with models but I was doing it for changes that affected all of miq_server and util.

This test would run the miq-password spec first, reset the key_root, not clear out the v*_key, and subsequent tests that run MiqDatabase.seed such as rhn_mirror_spec.rb would fail with seemingly cryptic errors like:

     Failure/Error: it("#decrypt")        { expect(MiqPassword.new.decrypt(enc_v2)).to           be_decrypted(pass) }
     OpenSSL::Cipher::CipherError:
       bad decrypt

Or

     Failure/Error: it("#recrypt legacy") { expect(MiqPassword.new.recrypt(enc_v0)).to eq(enc_v2) }

       expected: "v2:{27X41c6xqCCdVcw4LlQ1Qg==}"
            got: "v2:{kGuPkrN8fYq0cJPTjytisw==}"

These tests now don't randomly fail with this PR.

[miq-bot]

Checked commit jrafanie@6cbf5a4 with rubocop 0.33.0 and haml-lint 0.13.0
2 files checked, 0 offenses detected
Everything looks good. 🍰