fix: parse legacy LeaderHint from Oxia 0.16.x servers

[mattisonchao]

Motivation

The Java client only reads the shard-leader redirect from google.rpc.ErrorInfo metadata, which Oxia servers publish from 0.17.0. Servers on the 0.16.x line (including 0.16.8) attach the redirect as a proto.LeaderHint status detail instead. With a 0.8.0+ client against a 0.16.x server, the redirect is invisible to the client: retries keep targeting the client's cached (stale) shard leader and node is not leader for shard N errors persist until the client process restarts and re-fetches shard assignments. This was observed in production for ~6.5 hours after an Oxia pod replacement.

Modifications

• Add a LeaderHint message to the client proto, mirroring the 0.16.x server definition (shard, leader_address).
• Parse the legacy LeaderHint gRPC status detail in OxiaStatusException and surface it through the existing shard/leader metadata, so getLeaderHint() works against 0.16.x servers and retries follow the server-provided leader.
• Add a unit test covering the legacy 0.16.x wire format (type.googleapis.com/io.oxia.proto.v1.LeaderHint).

Testing

• ./gradlew :client:test --tests io.oxia.client.grpc.OxiaStatusExceptionTest --tests io.oxia.client.grpc.GrpcRpcProviderTest passes.
• :client:spotlessJavaCheck passes for the changed files.