[codex] stabilize pool scaling

[GodBlf]

Summary

• synchronize the Helm Pool CRD with PoolStatus.Updated
• make pool buffer calculations use Ready, unallocated pods while retaining Pending pods as in-flight capacity
• bound scale-down batches with scaleStrategy.maxUnavailable and wait for deletions to finish
• delete least-useful idle pods first and preserve stable Ready capacity
• count terminating pods against PoolMax
• keep status updates running while scale expectations are pending

Root cause

Six behaviors amplified one another under slow-starting runtimes: the Helm CRD pruned status.updated, buffer calculations treated Pending pods as available, the buffer target jumped to its midpoint outside the configured band, scale-down was unbounded, terminating pods disappeared from occupancy accounting, and scale-down selected the oldest Ready pods first. Together these caused a hot reconcile loop and sustained pod create/delete oscillation.

Impact

The controller now converges without repeatedly replacing usable pool capacity. Existing Pool manifests remain compatible: no CRD fields, CLI flags, or Helm values were added. The existing scaleStrategy.maxUnavailable setting now limits both scale-up and scale-down batch sizes; the default remains 25%.

Validation

• KUBEBUILDER_ASSETS="..." go test ./internal/controller -count=1 (36 envtest specs)
• targeted pool scaling stability unit tests
• go test ./internal/utils/expectations/... -count=1
• go test ./... -run '^$'
• go vet ./internal/controller ./internal/utils/expectations/...
• helm lint charts/opensandbox-controller
• rendered Helm CRD contains .status.updated and the UPDATED printer column
• pnpm docs:build
• git diff --check

Closes #1423

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Pangjiping]

LGTM. @fengcone @Spground ptal

[halleystar]

还没处理完？ 效率那么低吗？

[GodBlf]

修复 CI teardown hang：Core E2E 在所有测试完成后残留 BatchSandbox/Pool/SandboxSnapshot，make undeploy 删除 CRD 时因 finalizer 长时间等待而触发 1h timeout。现在 Core AfterAll 会在 controller 仍运行时按依赖顺序清理这些 CR，再执行 undeploy。

本地验证：make test、make lint、go vet ./...、Helm lint/template、pnpm docs:build 均通过。新提交的 Actions 目前显示 action_required，需要仓库维护者批准 fork PR workflow 后才会开始运行。

[GodBlf]

@Spground @fengcone @Generalwin @kevinlynx @ninan-nn @hittyt @jwx0925

Issue #1423 的 Pool 缩扩容修复已经完成，并已推送到当前 PR，最新提交为 01fb1d2a。另外已修复 Core E2E teardown：在 controller 仍运行时清理残留的 BatchSandbox、SandboxSnapshot 和 Pool，避免 make undeploy 删除 CRD 时被 finalizer 卡住。

本地已通过 make test、make lint、make helm-lint、go vet ./... 和 pnpm docs:build。PR 已同步到最新 main。当前新的 GitHub Actions 运行显示 action_required，请有仓库权限的维护者批准 fork workflow 后开始验证。

PR: #1425