refactor(cli): extract explore feishu sink commands into their own module

[NIU-123370]

Addresses task-board row GH-C06 ("CLI ownership and hot-module extraction", Available): characterize one remaining oversized CLI ownership seam and move one cohesive command group into its bounded module.

Characterization

The size/ownership smoke governs loopx/cli_commands/*.py only, so bootstrap_command_pack.py (tracked separately by the canary ratchet) is out of scope for this row. After the recent quota/status/todo/history/scheduler extractions, the remaining oversized seam under the smoke is loopx/cli_commands/explore.py at 989/1000 lines — the largest module in the directory and 11 lines from the default budget (runner-up: registry_admin.py at 975).

What moved

The four feishu-* Lark sink subcommands (feishu-setup, feishu-visual-configure, feishu-sync, feishu-card): their registration blocks, dispatch branches, the sync-only _target_config helper, and the 17 imports only they use — into a new loopx/cli_commands/explore_feishu_commands.py (320 lines), mirroring the existing explore_planning_commands.py sibling pattern: shared arg helpers injected as callables, no compatibility wrappers. Public invocations unchanged — loopx/cli.py still calls register_explore_commands / handle_explore_command, and explore --help / explore schema output is byte-identical to main.

• explore.py: 989 → 723 lines
• explore_feishu_commands.py: 320 lines (well under budget; no baseline entry needed, same as its sibling)

Validation (task-specified commands)

• python examples/cli-command-module-size-ownership-command-modularization-smoke.py → ok
• python regression/cli-command-module-contract.py → ok (requires pip install -e . for the loopx entry point and a LOOPX_REGISTRY fixture; the local-machine registry miss reproduces on main too — environmental)
• Focused pytest: tests/cli_commands/ + tests/canary/ → 81 passed; -k explore → 152 passed; examples/explore-feishu-singleflight-smoke.py (monkeypatch target updated to the new namespace) → ok

Two honest notes:

1. The size/ownership smoke is red on pristine main — refactor: reset benchmark research around native runners #3267 deleted two benchmark modules but left their LEGACY_MODULE_LIMITS entries, tripping the smoke's own stale-budgets assertion. This PR removes the two stale entries so the task's validation command is usable again (5 lines in the smoke script, called out here for review visibility).
2. tests/extensions/test_extension_scaffold.py has a collection error from a missing optional dependency that reproduces identically on base — pre-existing, untouched.

tests/cli_commands/, canary, and import-boundary suites stay green; mypy delta vs main is line-number drift plus one _SubParsersAction type-arg in the new register function matching the sibling modules' existing style.

Refs GH-C06.

[huangruiteng]

评审对象：59911d1e8daf8a7b72669c9e3fae80273025982d

动机

这次改动解决的是一个真实且已经接近阈值的维护性问题：loopx/cli_commands/explore.py 在主干上已达到 989 行，Feishu/Lark 展示 sink 的四个子命令又恰好形成一组边界清楚、共同依赖同一扩展 provider 的职责。把这组注册与执行逻辑移出 hot module，能让后续 Explore 核心命令和 Lark 展示侧分别演进，也避免继续用扩大行数预算来掩盖所有权混杂。

改动思路

整体方案是一次保持行为的内聚提取，而不是再造一层通用抽象：

• 新模块 explore_feishu_commands.py 直接拥有 feishu-setup、feishu-visual-configure、feishu-sync、feishu-card 的参数注册和执行分支；
• explore.py 只保留入口编排，通过 register_explore_feishu_commands 和 handle_explore_feishu_command 委派，并把已有的配置路径、projection 构造器和 runtime root 注入进去；
• FEISHU_SINK_EXPLORE_COMMANDS 同时驱动扩展激活检查和 dispatch，避免注册集合与执行集合各自漂移；
• 没有留下旧私有函数的兼容 wrapper，也没有把 Lark provider 细节提升成新的核心 capability，归属符合现有扩展边界。

具体改动

1. loopx/cli_commands/explore.py

   • 删除仅由 Feishu sink 使用的 17 个 import、四段 argparse 注册和四个执行分支；
   • register_explore_commands 改为调用 register_explore_feishu_commands；
   • handle_explore_command 在完成 registry/runtime/source-route/config-path 解析后，把 Feishu 子命令交给新 handler；外层仍统一追加 source_runtime_route、extension_activation，并统一把异常收敛成 CLI error packet。

2. loopx/cli_commands/explore_feishu_commands.py

   • register_explore_feishu_commands 完整保留四个公开子命令的参数、默认值、choices 和 help；
   • _target_config 继续按“显式参数优先、本地配置兜底”解析 Base/table/CLI identity，缺少完整目标时抛出明确错误；
   • handle_explore_feishu_command 保留 setup、visual configure、singleflight sync 和 card 文件/消息 ID 持久化四条路径；
   • 正向 feishu-sync 路径仍是：获取 singleflight → 构造 projection → 解析 target → 同步表格 → 同步单/多 visual sink → 汇总 ok；锁已占用时直接返回 not_attempted_sync_busy，明确标记未做外部写入。配置不完整或 provider 调用失败时，异常继续由外层 CLI handler fail closed。

3. 两个 smoke

   • singleflight smoke 只把 monkeypatch 目标迁到真实查找位置，继续证明锁外不误写、锁内按顺序同步 rows/visual；
   • module-size smoke 删除已经随 #3267 模块删除而失效的两个 legacy budget，恢复“预算不能指向不存在模块”的自检。该清理在 pristine main 上可复现红灯，因此与本次维护性任务相容。

对主干的风险

主要风险在于提取时遗漏 argparse 默认值、改变 dispatch/扩展激活顺序、破坏 singleflight 的 fail-closed 行为，或让 source-runtime 路由不再进入 projection。逐文件对照后未发现这些变化：四个命令仍共享同一个集合常量，projection callback 收到原 runtime root 和 override 参数，busy 分支仍不执行 rows/visual 写入，handler 返回后仍经过外层 route/activation 装饰与错误渲染。

本次没有新增或修改 typed state 规则；没有把产品/benchmark 文案写入通用 control-plane 合约；公开默认行为声明为 refactor，并由 help/schema parity 覆盖；也没有把机器义务误称为 guidance。

独立验证结果：

• git diff --check origin/main...HEAD：通过；
• CLI module-size smoke：通过；且 pristine main 的两个 stale legacy budget 失败可复现；
• Explore Feishu singleflight smoke：通过；
• CLI command module regression：通过；
• tests/cli_commands + tests/canary：81 passed；
• tests/architecture：13 passed；
• 变更文件 Ruff：通过；
• explore --help 与 explore schema 相对主干逐字节无差异；
• GitHub 上 DCO、dependency review、pytest、build 均为 green。

我的整体评价

APPROVE。 这是一次边界清楚、范围克制、可回滚的模块提取：它减少了 hot module 的所有权负担，同时保留了公开 CLI、扩展激活、source-runtime 路由、singleflight 和错误路径的语义。附带的 stale budget 清理有独立可复现依据，不是借机扩 scope。当前 exact head 未发现 blocker 或需要作者额外修改的 P2 项。

[huangruiteng]

APPROVE — reviewed exact head 59911d1e8daf8a7b72669c9e3fae80273025982d.

The extraction keeps the public Explore CLI contract and failure semantics intact: the four Feishu sink commands retain their arguments/defaults, extension activation and source-runtime routing stay in the outer handler, and feishu-sync still fails closed under the singleflight busy path. The stale module-budget cleanup is independently reproducible on pristine main.

Independent validation: diff check; both focused smokes; CLI command regression; 81 CLI/canary tests; 13 architecture tests; Ruff on all changed files; byte-identical explore --help and explore schema versus main; required GitHub checks green. No blocking findings.