Target-aware Deep Review reviewer scheduling

[limityan]

Summary

• Add target-aware Deep Review manifest construction, including frontend/domain classification, explicit /DeepReview target parsing, and git commit/range changed-file resolution.
• Persist the Deep Review run manifest through Flow Chat session metadata so continuation and child sessions keep the original reviewer scope.
• Enforce the persisted manifest in the backend Task pipeline so only active reviewers from the run manifest can be launched.
• Add custom review subagent tooling validation: require GetFileDiff + Read, share the tooling contract between UI and Review Team execution, and surface invalid reviewers as invalid_tooling skipped reviewers instead of silently dropping them.
• Add token-budget aware extra reviewer handling and skipped-reviewer reporting for disabled, unavailable, not-applicable, budget-limited, and invalid-tooling cases.

Verification

• pnpm --dir src/web-ui exec vitest run src/shared/services/reviewTeamService.test.ts src/app/scenes/agents/components/subagentEditorUtils.test.ts
• pnpm run type-check:web
• pnpm run lint:web
• pnpm --dir src/web-ui run test:run
• cargo test -p bitfun-core deep_review -- --nocapture
• git diff --check HEAD~4..HEAD