Support memory arrays in MerkleTree multiproof

[Amxx]

Fixes #3492

PR Checklist

• Tests
• Documentation
• Changelog entry

[frangio]

Why do we think the leaves are likely to be in memory?

[Amxx]

Most of the time, the leaves are a hash produced in by the contract. For example, you want to claim 100 tokens to for user X, you hash that (sometimes with a nonce)

Like shown in this contract of the merkledrop workshop.

Now if you want to claim multiple allocations at once, you would get an array of descriptions (number of tokens, recipient, ...), you would hash each element of the array, and that would be your leaves that need to be verified. It would be an array in memory.
The array in calldata would be the preimages.

You could imagine having the preimage AND the leaves in calldata, hashing all the preimages, and then checking that the hashes match the calldata array, but that doesn't sound right to me.

[frangio]

I've checked that the body of memory and calldata functions is the same.