feat: enhance prompt system with feature-conditional fragments and enriched tool descriptions

[yishuiliunian]

Summary

• Activate the dormant Condition::Feature mechanism and wire it from config state (memory, hooks, subagent, output style)
• Add 6 new prompt fragments: memory guidance, hooks explanation, sub-agent guidelines, anti-hallucination guardrails, and 2 output styles (explanatory, learning)
• Enrich tool descriptions for 8 tools (Read, Write, Edit, Fetch, WebSearch, EnterPlanMode, ExitPlanMode, AskUser) with behavioral guidance

Changes

Infrastructure (features wiring)

• system_prompt.rs: add features: Vec<String> parameter to build_system_prompt
• agent_setup.rs: build features vec from config (memory/hooks/subagent/style), extract MCP append to prompt_post.rs
• settings.rs: add output_style: String field

New prompt fragments (23→29)

• core/memory-guidance.md — when/how to use Memory tool
• core/hooks.md — treat hook feedback as user input
• tools/agent-guidelines.md — sub-agent spawning strategy
• tasks/anti-hallucination.md — no fabrication, verify before referencing
• styles/explanatory.md — educational response style
• styles/learning.md — collaborative learning style

Enhanced existing content

• security.md — expanded from 2 lines to 5-layer threat model
• memory-maintainer.md — 4-type taxonomy, frontmatter format, merge strategy

Tool descriptions enriched

• Read (image/PDF/notebook support), Write (read-before-write), Edit (indentation/uniqueness), Fetch (auth failure warning), WebSearch (sources requirement), PlanMode (when to use), AskUser (multiSelect/plan mode notes)

Test plan

• bazel build //... passes
• bazel build //... --config=clippy zero warnings
• bazel test //... — 49/49 pass
• Fragment count updated 23→29
• Feature-conditional tests (memory, hooks, subagent, style)
• New plan-mode tool tests (name, description, permission)
• Fetch description test added