feat(api-access): sync drag order into credential priority behind a toggle

[jiaxuan1101]

中文

背景

GUI 中拖动接入（Provider）排序目前只改变 config.yaml 里的数组顺序，而 CPA Core 调度时完全不使用这个顺序：选择器按 priority 分桶、只取最高可用档，档内按凭据的内容哈希 ID 排序（见 sdk/cliproxy/auth/selector.go）。因此拖拽排序对实际故障转移顺序没有影响。

改动

本 PR 通过核心已支持的 priority 字段让拖拽顺序生效：

• 在 API 接入页头部新增可选开关 "按列表顺序调度"（默认关闭）
• 开启后，拖动接入会将其列表位置写入该分类下各接入的 priority 字段（第一行优先级最高；末行 priority: 0 省略，因为 0 即核心默认值），核心在该分类内严格按从上到下的顺序故障转移
• 开启开关时立即按当前分类顺序同步一次优先级（同步时忽略搜索过滤，避免误伤未显示的接入）
• 关闭开关后行为与现状完全一致（拖动仅改写数组顺序），已写入的优先级保持不变
• 开关持久化到 GUI 配置的 ^Gpi-access-order-scheduling 字段（默认关闭）；#[serde(default)] 保证旧配置文件兼容

验证

• cargo check 与全量 cargo test：343/344 通过；唯一失败的 ^Ggent_command_path_prioritizes_the_executable_and_preserves_discovered_paths 在未修改的上游 main 上同样失败（该测试依赖本机 PATH 组成，与本改动无关）
• 前端 sc --noEmit 通过
• 已在 Windows 便携版实测：开启开关后拖动接入，优先级按位置写入并显示在列表行上；关闭后恢复原行为

说明

• 开启开关后，下次拖动或开启同步会覆盖该分类内手工设置的优先级；开关关闭时编辑对话框的手工 priority 字段照常生效
• 新增 en、zh-CN 与 ja 文案（项目现有三个语言包）

---

English

Background

Drag-to-reorder on the API Access page currently only rewrites the array order in config.yaml, which the CPA core scheduler ignores entirely: selection groups credentials into priority buckets and picks from the highest available tier, ordering within a tier by the stable content-hash auth ID (see sdk/cliproxy/auth/selector.go).

Changes

This PR makes the drag order meaningful through the mechanism the core already supports: the per-credential priority field.

• Adds an opt-in Schedule by list order switch on the API Access page header (off by default)
• When enabled, dragging a connection writes its list position into the priority field of that category's connections (first row gets the highest value; the last row's priority: 0 is omitted since 0 is the core default). The core then fails over strictly from top to bottom within the category
• Enabling the switch immediately syncs priorities from the current category order (the search filter is ignored for this sync)
• When disabled, behavior is unchanged (drag only rewrites the config array order), and already-written priorities are left untouched
• The toggle persists in the GUI config as ^Gpi-access-order-scheduling (default off); #[serde(default)] keeps older config files compatible

Validation

• cargo check and the full cargo test suite: 343/344 pass; the single failure (^Ggent_command_path_prioritizes_the_executable_and_preserves_discovered_paths) also fails on unmodified upstream/main on this machine (PATH-composition-dependent, unrelated to this change)
• Frontend sc --noEmit passes
• Tested manually on Windows with a portable build: toggling the switch and dragging connections writes priorities per position and displays them on the rows; disabling restores the previous behavior

Notes

• Once enabled, the next drag or toggle-on sync overwrites manually set priorities of that category; the edit dialog's manual priority field still works when the switch is off
• Labels added for en, zh-CN, and ja (the three locales the project now ships)

[jiaxuan1101]

顺便借这个 PR 提一个后续优化的想法（与本 PR 互不影响）：

当前的模型别名功能是把别名绑定到来源渠道，故障转移仍然只发生在同一 Provider 的多个凭据之间。如果能让每个模型单独配置一个故障转移列表（例如 claude-opus-4-5 → [凭据A, 凭据B, 凭据C]），没有显式配置的模型则默认沿用 Provider 级的故障转移顺序（比如本 PR 的"按列表顺序调度"），调度灵活性会再上一级——不同模型可以拥有各自独立的凭据优先链。

---

One related idea for a follow-up (independent of this PR): the current model-alias feature binds an alias to a source channel, and failover still happens only among credentials of the same provider. It would be nice if each model could have its own failover list (e.g. claude-opus-4-5 → [credential A, credential B, credential C]), while models without an explicit list keep the provider-level failover order (e.g. the schedule-by-list-order switch from this PR) as the default. That way different models can have their own independent credential chains.