Extend `EncodedChunk` abstraction to write side

Have
```py
Array.encode_chunk() -> EncodedChunk
```
which takes in chunk indices and a typed, decoded array (#47).

Or rather, because chunk encoding will be either sync (current thread) or async (on a Rust thread pool), maybe have a separate `ChunkEncoder` class.

Or, just put encoding onto `CodecChain`, so we have `CodecChain.encode` and `CodecChain.encode_async`.

- We probably want a way to encode/decode chunks that is independent of a store.
- It seems a little messy to put sync/async methods for chunk encoding onto `Array`/ `AsyncArray`, when the sync/async bridge of chunk encoding describes _where the CPU bound work is happening_, not what kind of IO is happening

Start by comparing the mentioned Array, AsyncArray, EncodedChunk, ChunkEncoder, and CodecChain abstractions and how chunk encoding should remain independent of a store. Resolve whether encoding belongs on a separate encoder or CodecChain, and define the sync and async API that satisfies the issue.

Written by the indexing model from the issue text.