feat: support device init once and reinit

[pfwang80s]

Context

data-platform now treats device init and reinit as separate operations:

• InitDevice: first-time device provisioning only.
• ReinitDevice: refresh credentials/config for an already-initialized device.

Related platform PR: archebase/data-platform#222

What Changed

• Added DeviceInitService.ReinitDevice to the SDK proto and regenerated Swift protobuf/gRPC bindings.
• Updated the Swift control-plane transport and ArchebaseDeviceInitializer so initDevice and reinitDevice call distinct RPCs.
• Kept the Qiongche public facade semantics unchanged: saveConfigAndInit(configString:) still performs init-or-reinit, falling back to ReinitDevice when the server reports DATA_GATEWAY_DEVICE_ALREADY_INITIALIZED.
• Updated Qiongche readiness so a successful remote init/reinit invalidates the previous local ready state until endpoints/config/state all commit again.
• Shared the init/reinit remote config fetch path between the generic initializer and the Qiongche provisioner.
• Updated README guidance and focused tests for explicit init/reinit behavior and Qiongche fallback/recovery behavior.
• Fixed iOS simulator smoke execution by preserving dependency Swift settings, passing simulator deployment target explicitly, propagating device-init integration env, normalizing simulator single-slash URLs, and failing early when the local init endpoint is missing.

Reviewer Follow-Up

• Generic ArchebaseDeviceInitializer.reinitDevice now works even when the local config is missing, so callers have an explicit recovery path after remote init succeeds but local persistence does not complete.
• Qiongche now clears its ready-state marker after remote init/reinit success and before local writes; isReadyToUpload() requires a valid state whose endpoint hash matches the endpoint file, preventing stale credentials from being reported ready after partial local persistence failure.
• Added a regression test proving Qiongche can recover after remote success plus local endpoint persistence failure, and that it reports not-ready during the failed intermediate state.
• Added early DGW_*_INIT_ENDPOINT validation to Scripts/simulator_smoke.sh local-mode execution and covered it with a script-shape test.
• Reverted the personal Xcode xcuserdata scheme-order change and added xcuserdata/ to .gitignore.
• Kept InitDeviceRequest / ReinitDeviceRequest as-is in this PR because changing that proto surface must be synchronized with data-platform; this is not safe to do SDK-only.

Validation

Latest reviewer-fix validation on commit b47add7:

• bash -n Scripts/simulator_smoke.sh
• swift test --filter DeviceInitializerTests：6 tests passed
• swift test --filter QiongcheDataGatewaySDKTests：32 tests passed
• swift test --filter LocalStackHarnessTests/simulatorSmokeScriptSkipsPackageUpdatesForCachedDependencies：1 test passed
• git diff --check
• swift test：222 tests passed; real/local gated integration tests skipped without their env

Earlier ACK validation for this branch before the reviewer follow-up:

• ACK SwiftPM direct real OSS e2e: 221 tests passed, log: .local/e2e/sdk-ack-20260520-4d6c8bc-203555/swift-direct-realoss.log
• ACK public-path SwiftPM real OSS e2e: 3 tests passed, log: .local/e2e/sdk-ack-20260520-4d6c8bc-203555/swift-public-path-realoss.log
• ACK iOS simulator smoke real OSS e2e: 3 selected tests passed, log: .local/e2e/sdk-ack-20260520-4d6c8bc-203555/simulator-smoke-realoss.log

Risks and Rollout

• SDK ReinitDevice calls require the companion data-platform server rollout; deploy after archebase/data-platform#222.
• Callers using the generic initializer should keep strict semantics: initDevice remains create-only, while reinitDevice is the explicit recovery/rotation path.
• Rollback by reverting this PR and continuing to use the previous SDK behavior.

Checklist

• PR title uses Conventional Commits.
• Tests and docs are updated or marked not applicable.
• Public Rust APIs have English documentation comments. N/A, Swift SDK repo.
• PR description matches the final implementation.

[zz-jason]

按前面约定的 review 口径提交 inline comments：重点看了 init/reinit 状态机、proto 设计、脚本验证路径，以及是否有过度包装/可精简的重复实现。主要风险是通用初始化路径在“服务端已 init、但本地配置丢失/未落盘”时缺少恢复入口。

[zz-jason]

Review summary

这次按前面约定的口径看了 correctness、contract/proto、测试证明、脚本验证路径，以及是否有过度设计或可精简代码。

主要结论：

1. 需要优先修复恢复路径：通用 ArchebaseDeviceInitializer 在“服务端 InitDevice 已成功，但响应丢失/本地配置未落盘/本地配置被删”后会卡住：再次 init 得到 already initialized，reinit 又要求本地配置存在。建议允许本地缺失时显式 reinit，或在本地缺失且远端 already initialized 时 fallback 到 ReinitDevice，并补回归测试。
2. 穹彻 init-or-reinit 需要补失败恢复证明：saveConfigAndInit 的 reinit 会轮换远端 credential，但本地 endpoint/config/state 是后续分步写入。建议补 reinit 成功后本地任一步写入失败的恢复测试，证明再次提交配置可以恢复，且失败期间不会把旧 credential 误判为 ready。
3. proto 可以精简：ReinitDeviceRequest 与 InitDeviceRequest 字段完全一致。若没有明确未来差异，建议复用 request message 或改成共享 DeviceInitRequest，避免生成代码膨胀和后续契约漂移。这个需要和 data-platform 侧 proto 同步调整。
4. 脚本验证入口需要更早失败：simulator smoke 默认会跑 device-init 流程，但本地模式入口没有强校验 DGW_LOCAL_INIT_ENDPOINT。建议在脚本入口就校验，避免拖到 xcodebuild 测试阶段才失败。
5. 代码精简点：FilePreparation.swift 和 QiongcheDeviceProvisioner.swift 各自维护了一套 remoteConfig + DeviceInitRemoteMode，建议抽 package-private helper 复用，减少错误映射、sdkVersion/platform、response 校验后续漂移。
6. repo hygiene：PR 带了个人 Xcode xcuserdata scheme 排序变更，建议移除，并考虑忽略 xcuserdata/。

补充观察：SDK 的设备 init/reinit 路径没有被通用 RetryExecutor 自动重试包住；这次主要风险不是 retry executor，而是一次性 init 成功后 SDK 本地无配置时缺少恢复入口。