NDArray/multi-columnar with efficient CRUD operations?

I'm in the hunt for material or implementations of NDArray and/or columnar structures with not-bad support for CRUD operations (insert-update-delete) that could work in-memory.

I'm aware of kdb+ but my understanding is only for big append-only loads then calculations on it.

Currently, I have for my own little relational lang (http://tablam.org) several structures that backed the relations (BTreeMaps - Vectors - Table - Scalars). The main is a table:

https://bitbucket.org/tablam/tablam/src/41b3b0676d6062031998af48b683f8f0062bf279/core/src/types.rs#lines-221

#[derive(Debug, Clone, PartialEq, PartialOrd, Eq, Ord, Hash)]
pub struct Table {
pub schema: Schema,
pub data: Vec,
}

and this is ok. But wonder what other options I could explore. (The allure for me in use just a NDArray-like container is that I can reduce my implementations to just 2 BtreeMaps - NDArray)