[rust] Add RPC message wrappers for core DB/table operations

[gnuhpc]

What

Adds request/response message wrappers for the core 1.x DB/table admin RPCs: AlterClusterConfigs, AlterDatabase, AlterTable, CreateAcls, DescribeClusterConfigs, DropAcls, GetTableStats, ListAcls, ListDatabaseSummaries.

Each is a thin wrapper over the generated proto:: types (from #628) wired into rpc/message/mod.rs. The admin client methods that consume them land in a later PR.

Stack

Part 2/6 of the Fluss 1.x Rust support series, stacked on #628 (pr/1-proto-sync). Each branch in the stack compiles on top of its predecessor, so all PRs here target main; once #628 merges, this PR's diff automatically reduces to just the 9 message files unique to this branch.

🤖 Generated with Claude Code

[fresh-borzoni]

@gnuhpc Thank you for the PR, we need to address a couple of issues here.

We shouldn't use proto::* in any public FlussAdmin signature.
Please, define domain types mirroring Java's with conversion in the wrapper layer

[fresh-borzoni]

why do we hardcode target_columns: vec![]?
There's no way to ask for specific columns with this, is it intentional?

[gnuhpc]

Not intentional — fixed in 720d069. GetTableStatsRequest::new now takes target_columns: Vec<i32> and threads it through. Single existing call site (the integration test on #632) will pass an explicit vec![] to preserve current behavior. Same drop-silently-features pattern also fixed on #631 (alter_table was hardcoding vec![] for drop/rename/modify columns); will be in the next force-push on that PR.

[fresh-borzoni]

Can we drop #[derive(Default)] on the arg-taking requests?

default() just builds empty(invalid requests), and the rest of the wrappers are #[derive(Debug)] only, so it's inconsistent

[gnuhpc]

Done in 720d069. Dropped #[derive(Default)] on all 7 arg-taking wrappers added by this PR (alter_cluster_configs, alter_database, alter_table, create_acls, drop_acls, get_table_stats, list_acls).

Kept Default on the param-less ones (describe_cluster_configs, list_database_summaries) since they match the pre-existing list_databases.rs precedent and default() produces a valid request there — happy to drop those too if you'd prefer strict uniformity.

Same change applied to the 18 arg-taking wrappers on #630.

[gnuhpc]

Addressed the proto::*-in-public-API feedback in 720d069. Summary of what changed:

Domain types introduced under crates/fluss/src/metadata/ (mirroring Java with to_pb / from_pb conversion in the wrapper layer):

• config.rs — AlterConfig + AlterConfigOpType (Set/Delete/Append/Subtract)
• acl.rs — AclInfo / AclFilter + ResourceType (Any/Cluster/Database/Table), OperationType (Any/All/Read/Write/Create/Drop/Alter/Describe), PermissionType (Any/Allow)
• table_change.rs — AddColumn / DropColumn / RenameColumn / ModifyColumn + ColumnPositionType (currently Last only, matching Java's ColumnPositionType.java — proto reserves First=1, After=3 for future use)
• table_stats.rs — BucketStatsRequest

Enum int codes are taken verbatim from the Java source (AlterConfigOpType.java, ResourceType.java, OperationType.java, PermissionType.java, ColumnPositionType.java).

Wrapper changes: all 7 arg-taking wrappers' new() now accept domain types and call .to_pb() to populate inner_request. Wrappers still hold proto::* internally (matches the existing PartitionSpec/DatabaseDescriptor pattern).

Stack-wide sweep: since the same proto::*-leak class of issue exists on #630 and #631 (the FlussAdmin admin client surface in #631 was where most leaks lived), I'm restacking the rest of the chain with the same cleanup now — will post here once each PR is force-pushed.

Tests: added *_pb_roundtrip unit tests for each new domain type (matching the PartitionSpec/DatabaseDescriptor test style). cargo test -p fluss-rs --lib passes 544 (15 new).

Thanks for the review!

[fresh-borzoni]

@gnuhpc Thank you for the changes, LGTM
I pushed commit to tighten visibility for protos in your new code and a couple of pre-existing issues. Hope it's okay with you