[Flow EVM] Optimize speed and memory usage of byte slice conversions in Cadence

[fxamacker]

Updates #8401

NOTE: Although this requires modifying code in onflow/atree, onflow/cadence, and onflow/flow-go repos, the storage format is unchanged.

I think we can significantly optimize array and byte slice conversions by doing the conversions in onflow/atree rather than client software such as onflow/cadence and onflow/flow-go.

Using the new functions provided by Atree will optimize various operations such as decoding RLP data and calling EVM functions.

Some conversions can be roughly 10x faster while also reducing memory by 8x (bytes/op and allocs/op).