Use cheaper hash function for computing the DataKey

[joshua-spacetime]

We current use Keccak256 for computing a product value's DataKey. This can account for over 90% of the cost of generating the DataKey in some cases, the other portion being the bsatn encoding. Note this hash is invoked once for each row, both on the write path as well as the read path.

Is a cryptographic hash necessary for identifying rows? Can we use something cheaper?

[joshua-spacetime]

The DataKey computation also accounts for nearly 50% of the cost of query and subscription evaluation.

[jdetter]

Closed in favor of the memarch refactor

Blocker for fully removing the data key is the SDKs (right now they use data key for identifying rows for deletion). The SDKs should just use the row itself.