feat: smart testing mode — skip Playwright for API/backend features

[maswa]

What this does

Adds a "Browser Testing" setting with two modes:

• Full (default): Playwright runs for every feature, same as today
• Smart: Playwright is only loaded for UI features. API/backend/database features skip it entirely.

This saves ~100-300MB of memory per agent when testing features that don't need a browser, which adds up quickly in parallel mode.

How it works

A new should_use_playwright() function in client.py checks the feature's category against a list of backend keywords (api, backend, database, db, server, endpoint, service). If the category matches, Playwright MCP server and tools are excluded from that agent session.

The setting flows through the full stack:

• Settings UI → REST API → process manager → agent subprocess → client configuration

Changes

Backend (8 files):

• client.py — should_use_playwright() function, conditional Playwright inclusion
• agent.py — Feature category lookup from DB when in smart mode
• autonomous_agent_demo.py — --testing-mode CLI flag
• parallel_orchestrator.py — Pass testing_mode to subprocess commands
• server/routers/agent.py — Read testing_mode from settings, pass to manager
• server/routers/settings.py — Persist testing_mode setting
• server/schemas.py — Validation, request/response schemas
• server/services/process_manager.py — Store and forward testing_mode

Frontend (3 files):

• SettingsModal.tsx — Full/Smart toggle buttons (disabled when YOLO is on)
• useProjects.ts — Default setting
• types.ts — TypeScript types

Test plan

• Toggle between Full and Smart in Settings, start an agent, verify the correct mode is logged
• In Smart mode: API feature should log "NO Playwright (testing_mode=smart, category=api)"
• In Smart mode: UI feature should still get Playwright
• YOLO mode should override Smart mode (no Playwright regardless)
• Parallel mode should pass testing_mode to each subprocess