Fix simple-keyvalue-client crash

[kostko]

Fixes #2531

The crash reported in the above issue is actually result of two separate problems, both of which should be fixed by this PR:

• The runtime client endpoint should return empty sequences instead of nil as serde doesn't know how to decode a NULL when the expected type is a sequence.
• The simple-keyvalue client used a hardcoded round number in the query (3) which can easily be incorrect due to the time at which the client is called.

[codecov]

Codecov Report

Merging #2542 into master will increase coverage by 0.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #2542      +/-   ##
==========================================
+ Coverage   66.95%   66.96%   +0.01%     
==========================================
  Files         328      328              
  Lines       30184    30184              
==========================================
+ Hits        20210    20214       +4     
- Misses       7455     7462       +7     
+ Partials     2519     2508      -11

Impacted Files	Coverage Δ
go/runtime/client/client.go	69.35% <100%> (ø)	⬆️
go/genesis/api/api.go	26.19% <0%> (-7.15%)	⬇️
go/worker/common/p2p/p2p.go	62.61% <0%> (-4.96%)	⬇️
go/consensus/tendermint/abci/mux.go	63.4% <0%> (-3.99%)	⬇️
go/oasis-node/cmd/node/node.go	53.8% <0%> (-3.34%)	⬇️
go/oasis-node/cmd/debug/storage/storage.go	44.44% <0%> (-3.34%)	⬇️
go/runtime/transaction/transaction.go	76.82% <0%> (-1.22%)	⬇️
go/storage/client/client.go	75.92% <0%> (-0.93%)	⬇️
go/consensus/tendermint/tendermint.go	62.5% <0%> (-0.87%)	⬇️
go/worker/common/committee/group.go	81.78% <0%> (-0.75%)	⬇️
... and 9 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 83df083...71e63ec. Read the comment docs.

[mitjat]

Nice finds! Thank you!

Sucks that golang didn't take a clue from Java about how painful it is to maintain nil-safety.

[kostko]

Sucks that golang didn't take a clue from Java about how painful it is to maintain nil-safety.

In this case the returned nil is not unsafe. The problem is that in Golang both a nil and an empty slice can be used interchangeably (e.g., in range constructs and append etc. builtins). But when serialized into CBOR a nil is encoded as CBOR NULL while an empty slice is encoded as an empty sequence.

However the Rust serialization library (serde) doesn't know how to deserialize a sequence (e.g. Vec<T>) from a NULL by default and decoding fails.