feat: support non-blocking shard actions

[leon-ape]

What changed

• restrict Action.nonBlocking to explicitly supported API fields and whitelist shardAdd / shardRemove
• persist the selected shard/Pod target snapshot for non-blocking shard actions and reuse it across reconciles
• poll every selected target, aggregate results, and rerun only terminally failed Pods
• serialize overlapping requests by (Pod, Action) while allowing disjoint requests to progress concurrently
• preserve the existing shardAdd/shardRemove business gates and the blocking path
• regenerate CRDs, Helm CRDs, and API documentation

Why

Shard add/remove actions can run for minutes while moving data or rebalancing a cluster. The existing blocking controller path cannot reliably handle those actions within the synchronous call limit. This change lets the two sharding actions opt into the non-blocking Action framework introduced by #10715 without enabling non-blocking mode for unrelated lifecycle actions.

User impact

Addon authors may set nonBlocking: true on ShardingDefinition.spec.lifecycleActions.shardAdd or shardRemove. Existing definitions remain blocking by default. This PR does not change any addon configuration.

Validation

• make generate
• make manifests
• make doc
• go test ./controllers/apps -count=1 with envtest assets
• go test ./controllers/apps/cluster -count=1 with envtest assets
• go test ./pkg/controller/lifecycle -count=1 with envtest assets
• go test ./pkg/kbagent/... -count=1
• go vet ./controllers/apps/cluster ./controllers/apps ./pkg/controller/lifecycle
• make golangci-lint (0 issues)

[apecloud-bot]

Auto Cherry-pick Instructions

Usage:
  - /nopick: Not auto cherry-pick when PR merged.
  - /pick: release-x.x [release-x.x]: Auto cherry-pick to the specified branch when PR merged.

Example:
  - /nopick
  - /pick release-1.1

CLA Recheck Instructions

Usage:
  - /recheck-cla: Trigger a re-check of CLA status for this pull request.
Example:
  - /recheck-cla

[khalildai]

CI 巡检分型，绑定 current exact 2b09d92cbfa752ac5d507324b6509004b8ef8f4a：

• push-pre-check (test) run 30517373526 / job 90790169371 的 dataprotection 套件执行了 153/153 specs，152 passed；唯一失败发生在 Volume Populator Controller 的 AfterEach 清理，不是主体断言。
• 清理在删除 Restore kb-populate-e3e70ae0-083d-4400-af41-046c59e5527e 时等待 DeletionTimestamp != nil，对象始终保留 nil，最终由 Gomega async assertion panic。
• 失败摘要：inferences source target from PVC labels for multi-target backups 主体路径完成后，cleanup panic；建议先检查 delete 请求/客户端错误是否被吞、finalizer/ownerReference 与 envtest delete semantics，并让 cleanup 在超时前输出 delete error 与对象 finalizers。
• 另一个 PR Pre-Check failure 是未关联 issue 的治理门，与上述 152/153 测试结果分开处理。

本次只补 current-head 证据，未 rerun/cancel。

[weicao]

The up-to-date fast path ignores the persisted non-blocking shard-action annotations. After target selection is saved and the Component and Cluster generations catch up, delayed and watch-driven reconciles exit before invoking or polling shardAdd, leaving the shard transition pending while Cluster status reports the desired generation as observed.

Reversing scale-out while a non-blocking shardAdd is running on an AllShards target that survives the reversal makes handleShardRemove return without polling or cancelling the detached add, and the source shard is then deleted. The obsolete action can continue changing membership or data on surviving shards after its source is removed, leaving the shard topology inconsistent.

[codecov]

Codecov Report

❌ Patch coverage is 69.07407% with 167 lines in your changes missing coverage. Please review.
✅ Project coverage is 64.69%. Comparing base (9ee2a73) to head (631dd49).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
...ollers/apps/cluster/sharding_nonblocking_action.go	65.82%	56 Missing and 25 partials ⚠️
...llers/apps/cluster/sharding_nonblocking_handler.go	64.42%	65 Missing and 9 partials ⚠️
...lers/apps/cluster/transformer_cluster_component.go	57.14%	4 Missing and 2 partials ⚠️
...ps/cluster/transformer_cluster_component_status.go	0.00%	2 Missing and 1 partial ⚠️
...apps/component/transformer_component_validation.go	66.66%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #10725      +/-   ##
==========================================
+ Coverage   64.56%   64.69%   +0.13%     
==========================================
  Files         521      507      -14     
  Lines       63343    63716     +373     
==========================================
+ Hits        40897    41223     +326     
- Misses      18804    18820      +16     
- Partials     3642     3673      +31     

Flag	Coverage Δ
unittests	64.69% <69.07%> (+0.13%)	⬆️

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.

[weicao]

Exact-head dependency recheck at 631dd49b4b63610fd35b3e17c849548ce645bcae: CI is green, but the 2,048-shard persistence blocker still applies.

The new 441dbac commit isolates orchestration into sharding_nonblocking_handler.go; it does not change target-state storage. selectShardingActionTargets still enumerates every selected shard/Pod, and setShardingActionTargets still serializes the complete raw JSON into one source Component annotation. The existing supported 2,048 shards x 3 Pods, All/All snapshot remains 391,193 bytes, above Kubernetes' 262,144-byte total annotation limit. There is still no size guard, alternate storage path, or boundary test.

Therefore Kubernetes can reject the first target-persistence update before any shard action is dispatched. Please move/shard/bound this state and add an exact 2,048-by-3 regression. This head change invalidates prior-head runtime evidence but does not alter the addon source fix in kubeblocks-addons#3285.