fix: improve key handling

[AaronFeickert]

Description

Updates the use of CoreTransactionAEADKey and CommsNoiseKey to be populated in place. Switches CommsNoiseKey to use a SafeArray under the hood.

Extends the work of PR 4967.

Motivation and Context

Earlier work updates the use of keys for value encryption and Noise. Value encryption keys use a SafeArray type in a Hidden wrapper, and Noise keys use a Hidden array. However, in both cases, a copy of the hash output used to populate the keys is left in memory.

This work mitigates the problem. Because the hashing API now supports in-place output finalization via finalize_into and finalize_into_reset, we can populate keys directly by mutable reference.

It also renames CoreTransactionAEADKey to EncryptedValueKey for clarity, and switches CommsNoiseKey to be a SafeArray under the hood. There is discussion on whether CommsNoiseKey needs to be a SafeArray; while the reasoning makes sense, I still think it's good practice to take advantage of the benefits of SafeArray for array-like key types unless there's a compelling performance reason otherwise. Discussion on this is welcome!

How Has This Been Tested?

Existing tests pass.

[AaronFeickert]

Before merging, it should be decided whether or not the domain separation labels for the relevant hashers should be updated to reflect the reverse-domain notation used elsewhere in the codebase. I think they probably should, but this will be a breaking change.