[keymgr/kmac] how kmac should send `kmac_data_i.error`

As brought up in this comment, we may want to find a better way to send the error when KMAC detects a fault during an operation requested by keymgr. We haven't had a conclusion in that discussion, so file this issue to continue on it.

Here are some approaches we brought up.

1. current implementation (no design change needed): kmac sends error without done, and keymgr gets stuck at the operation. Eventually, the chip will be reset as kmac sends a fatal alert.
2. make the operation complete normally: change Kmac also send done along with error. Also need to set ready high to accept the remaining data from keymgr.
3. treat the error as an interrupt to keymgr: keymgr finishes the operation once it sees the error is set. This changes the handshaking protocol, which impacts the current DV. My assumption was that error only asserts when done is set. We need to update the block-level TB if we choose this.

As a DV person, slightly prefer to 1) or 2) as they have less DV impact. I'm open to 3) as well, if it makes more sense to the design.

cc: @ballifatih @tjaychen @eunchan @cindychip

Read the linked discussion in pull request 16661 and inspect the current KMAC/keymgr handshake around kmac_data_i.error, done, and ready. Compare the three proposed behaviors and check the block-level DV impact; the work is done only after the protocol choice is agreed and the affected DV coverage is updated.