DirectKMSMaterialProvider API Behavior/Design improvements

[lavaleri]

Problem:

The correct behavior of the DirectKMSMaterialProvider is to use the configured CMK on encrypt, and to let KMS determine the correct CMK to use on Decrypt (this is a common KMS pattern, as KMS ciphertext stores the CMK used as metadata).

However, this behavior for DirectKMSMaterialProvider is potentially confusing, as customers may expect that the CMK configured on the CMP is also "used" to decrypt, and may be surprised if decryption succeeds even though the configured CMK was not the CMK used to encrypt the data.

Solution:

Since the original DirectKMSMaterialProvider was designed, KMS has introduced a keyId param on Decrypt that ensures the call fails if a different key was used to encrypt the ciphertext.

We should consider either updating or replacing the DirectKMSMaterialProvider to allow enforcing a particular key on decrypt, similar to the Strict vs. Discovery modes expressed by the AWS Encryption SDK's Keyrings and Master Key Providers.

Our new design should:

• maintain API parity between the DDBEC for Python and Java.
• be as simple as possible to reason about.
• minimize possible "modes" for behavior, and ensure that any "mode" needs to be explicitly chosen by customers on config.
• ensure that any default configuration/behavior chooses the safest/most conservative option for customers.

[lavaleri]

See related Python: https://github.com/aws/aws-dynamodb-encryption-python/issues/176