Replacement of KeyValueStoreView by a direct version.

[MathieuDutSik]

Motivation

The existing KeyValueStoreView is a scalable design for implementing a Key-Value store, which is needed by the EVM smart contracts and the Wasm smart contracts. However, it is a slow design that is overkill for 99% of the use cases.

Replacing it with an enum Big/Small might be a solution but it presents an inconvenient that we cannot easily transition from a Big sdesign to a small design if the storage use decreases. Also:

• It creates a strong discontinuity in the usage performance near the chosen threshold.
• Anyway, when we compute the hash, we have to access the whole set of key/values of the KeyValueStoreView.

Proposal

The following is done:

• The whole Key/value store is replaced by a single BTreeMap<Vec<u8>, vec<u8>>.
• Therefore, the API becomes a RegisterView, and this container is chosen accordingly. Note that this doubles the memory usage, but this was already the case for the small container in the now-closed PR.
• We cannot use the derive macro in linera-views. Therefore, the file key_value_store_view.rs is moved to linera-execution.
• The ViewContainer is eliminated and with it the corresponding tests.

Test Plan

The CI was adequate for identifying problems with the design.

Release Plan

• Nothing to do / These changes follow the usual release cycle.

Links

None.

[ma2bd]

The existing KeyValueStoreView is a scalable design for implementing a Key-Value store

Except that hashing is going to force reading everything anyway

[ma2bd]

This doesn't belong here, does it? Especially if we want to keep the unit tests.

[ma2bd]

Also it could be called SmallKeyValueStoreView if we want to keep the old implementation around for performance comparison (which are not done in this PR)

[ma2bd]

As discussed, we'd like to

• keep the old KVSView in place
• have the new "small" data-structure in linera-view as well (if possible -- otherwise maybe that's ok for now).
• make the change in linera-execution later, when we have benchmarks