fix(review): restore repository and launch error handling

[guantw]

Summary

• Preserve Git repository ownership failures as git_repository_untrusted and show localized, actionable errors in the Review platform panel. Automatic snapshot/context loads only display the error; explicit refresh and retry actions use the existing consent-based trust recovery flow.
• Share action-start error presentation between the Review action header and notifications. Prefer existing structured launch message keys, retain the dialog-start prefix as a compatibility fallback, preserve backend reasons, and let multiline details wrap.
• Distinguish missing local explicit file/directory targets from existing targets without workspace changes. Bind absolute paths to the workspace and use Git change evidence first so deleted files and rename source paths remain reviewable. Remote workspace Review keeps its existing unsupported result without probing paths or falling back to local inspection.

Related: #2746.

Type and Areas

Type: Regression fix.

Areas: Rust integration services, Core compatibility facade, desktop/Tauri commands, Web UI Review panels and launch flow, and en-US/zh-CN/zh-TW locales.

Motivation / Impact

Repository ownership failures otherwise surface as raw Git diagnostics, action-start errors can show inconsistent untranslated text across the header and notification, and nonexistent explicit targets lack useful feedback. The updated behavior makes these failures actionable while keeping trust prompts tied to explicit panel actions and preserving the current remote Review capability boundary.

Verification

• Focused Web UI suite: 151 tests passed across 12 files using pnpm --dir src/web-ui exec vitest run with --maxWorkers=50%. Coverage includes Git trust recovery/presentation, Review platform linking and mounted-panel trust interactions, action-bar formatting/rendering/i18n/layout, launch errors, explicit-target resolution, ReviewService, and SystemAPI.
• New interaction regressions cover automatic snapshot/context loads without trust prompts, explicit Retry approval/decline and bounded replay, identical header/notification errors, structured launch keys and legacy strings, local deletion/rename evidence, and remote exists/missing/offline cases with no path or Git probes.
• pnpm run check:web: passed, including type checking, Canvas SDK, typography, Appearance, theme color, and visual contract checks.
• pnpm run i18n:audit: passed with zero warnings.
• Backend verification for the unchanged Rust changes in this PR:
  • cargo test --locked -p openbitfun-services-integrations --no-default-features --features review-platform --lib repository_trust: 4 passed.
  • cargo test --locked -p openbitfun-services-integrations --no-default-features --features review-platform --lib ordinary_git_failure_remains_an_invalid_repository_error: 1 passed.
  • cargo test --locked -p openbitfun-services-integrations --no-default-features --features git --lib git::trust::tests: 19 passed.
  • cargo check --locked -p openbitfun-core --no-default-features --features review-platform: passed.
  • Desktop test binary compiled and linked with cargo test --locked -p openbitfun-desktop --lib api::review_platform_api::tests. With the test-artifact workaround below, api::review_platform_api::tests passed 4 tests and path_probe_payloads_accept_optional_remote_scope passed 1 test.
  • pnpm run check:core-boundaries: passed.
• git diff origin/main...HEAD --check: passed.

Reviewer Notes

• No persisted shape or backend protocol changes. Existing string-only action errors remain readable; structured launch errors use the existing launch error contract. The shared Git trust helper retains its existing behavior for other callers.
• Windows desktop tests required embedding Tauri's stock Common Controls v6 manifest into the generated test executable. Without it, the executable exits with STATUS_ENTRYPOINT_NOT_FOUND on this machine. This workaround affects only the test artifact and is not included in the PR. Existing matching sherpa-onnx 1.13.4 libraries were used for linking.
• Mounted React interaction tests and mocked API/runtime tests were exercised locally. Remote workspace rejection was verified with mocks; live SSH, remote control, Peer Device Mode, and Detached Dispatch were not exercised. This PR does not add remote workspace Review support.
• AI-assisted implementation with focused automated verification; no end-to-end application UI verification.

Checklist

• This PR is focused and does not include secrets, temporary prompts, generated scratch files, or unrelated artifacts.
• Relevant verification is recorded above, or skipped checks are explained.
• User-facing strings, docs, and locales are updated where applicable.