deserialization performance

[quininer]

I submitted djkoloski/rust_serialization_benchmark#56 and found that the performance of cbor4ii on github actions is worse than serde_cbor.

I also observed that cbor4ii has a greater slowdown than serde_cbor when using cranelift backend. (10x)

I suspect this is because cbor4ii relies more on inline than other implementations. This is a case that can be optimized.

[quininer]

Right now we are the fastest CBOR crate for serialization, but deserialization has no advantage over serde_cbor.