feat: add Set container type

[sshaplygin]

Closes #234.

Encoding

YDB has no dedicated wire type for sets. Per the container types documentation, Set<T> is "a special case of a dictionary with the Void value type", and ydb-go-sdk implements it that way — Set.ToYDB emits a DictType with a void payload, and setValue.toYDB emits pairs whose payload is the void value. This mirrors that encoding.

The raw layer already had RawType::Dict and RawValue::Pairs, so only the public mapping was missing.

What's added

• Value::Set(Box<ValueSet>) with Value::set_from(example, members), validating members against the example exactly like list_from.
• HashSet<T> ↔ Set<T> in both directions — the set analogue of Vec<T> ↔ List<T>. FromIterator for Value already claims iterators for List, so sets convert through From on the concrete HashSet.
• A Set section in examples/container-types.rs.

Two deliberate differences from the Go SDK:

• Empty sets keep their element type. Go falls back to EmptySet/EmptyDict because it has no type information when the item list is empty; set_from always has the example, so it emits Dict<T, Void> with zero pairs, which round-trips.
• A dict with a non-void payload is rejected on decode rather than read as a set with the payloads dropped. Value has no Dict variant yet (feat: Containers (Struct Variant List Tuple) #309), so this fails loudly.

Members are sent in the order given and are not de-duplicated, matching Go. YDB requires unique dict keys, so duplicates are the caller's responsibility — stated in the doc comment. Happy to add validation if you would prefer it over the O(n²) cost.

Two decoding bugs this surfaced

The unit tests passed, but running against a real server did not. TryFrom<ProtoValue> for RawValue rejected any Value message carrying no oneof, no items and no pairs — which is exactly how the wire spells two legitimate things:

1. Void. It has no representation of its own, so it arrives as the payload of every pair in a Dict<T, Void>. Fixed in the pair conversion, where the payload is known to be void.
2. An empty container. An empty list or dict carries neither items nor pairs, so it decoded as "empty value item". The generic decoder now yields NullFlag and lets the type-directed layer build the right empty value — which is what the existing (RawType::List(_), RawValue::NullFlag) arm was already written to expect.

The second is not new with Set: reading back an empty List<T> fails the same way on master today. empty_list_roundtrips_through_the_server covers it. Both new integration tests fail if the fix is reverted.

Verification

Run against ydbplatform/local-ydb:nightly — the image CI uses:

cargo test --workspace -- --include-ignored     # 314 passed, 0 failed
cargo fmt --check
cargo clippy --workspace --all-targets --no-deps --exclude=ydb-grpc -- -D warnings
cargo check -p ydb --features force-exhaustive-all --all-targets

Master on the same image is 302 passed / 0 failed, so nothing regressed.

Server-side coverage: Set<Int32> round-trip, DictContains (which only type-checks if the server really sees a dict/set), an empty set, and a Set<Utf8>. Unit coverage: the wire shape (dict type, void payload, null-flag pairs), rejection of a dict with a real payload, the set_from validation paths, and the HashSet conversions including a mismatched element type. Adding the variant also extends the Value::COUNT fixture, so Set now flows through the existing value/proto round-trip test.

Note for reviewers

Adding a variant is source-compatible under the default non_exhaustive, but breaking under the force-exhaustive-all feature — which is what that feature exists to surface. It compiles clean either way.

Unrelated but worth a separate issue: docker-compose.yaml pins ydbplatform/local-ydb:latest, while CI uses :nightly. On :latest, 11 of the existing integration tests fail with Unknown name: $val — every query-service parameter test. Local contributors following the README hit that before touching any code.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 87.00000% with 39 lines in your changes missing coverage. Please review.
✅ Project coverage is 86.98%. Comparing base (a6d7911) to head (972e4a2).

Files with missing lines	Patch %	Lines
ydb/src/types_test.rs	89.33%	16 Missing ⚠️
.../grpc_wrapper/raw_table_service/value/value_ydb.rs	75.55%	11 Missing ⚠️
ydb/src/types.rs	78.78%	7 Missing ⚠️
..._wrapper/raw_table_service/value/value_ydb_test.rs	90.00%	4 Missing ⚠️
.../src/grpc_wrapper/raw_table_service/value/proto.rs	85.71%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #593      +/-   ##
==========================================
+ Coverage   86.91%   86.98%   +0.07%     
==========================================
  Files         198      198              
  Lines       19492    19791     +299     
==========================================
+ Hits        16941    17216     +275     
- Misses       2551     2575      +24     

Flag	Coverage Δ
rust-1.88.0	86.97% <87.00%> (+0.06%)	⬆️
rust-1.96.1	87.31% <89.07%> (+0.15%)	⬆️
tests	86.98% <87.00%> (+0.07%)	⬆️
ubuntu	86.98% <87.00%> (+0.07%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.