Because log_filter_tree uses SSZ merkleization which uses SHA2. I’m not sure whether and when the existing EL structures will be converted to SSZ but for this purpose a binary tree is absolutely the logical choice, Merkle multiproofs are nice and easy with SSZ hash trees, so I think that sticking with the old hexary MPT format would be really impractical. And if we use an existing format with existing tooling then I don’t see a reason to complicate that by not using the hash function it was specified with.