fix: Improve subagent implementation-approach.md references for better AI execution accuracy

[shinpr]

Summary

This PR improves how subagents reference and utilize the
implementation-approach.md file, ensuring proper implementation strategy selection
and progress tracking across both Japanese and English environments.

Problem

• Subagents (technical-designer, work-planner) were not referencing
  implementation-approach.md, resulting in missing vertical/horizontal slice
  considerations
• task-executor's progress update instructions were too abstract ("3-location
  synchronized update"), leading to checkbox update omissions
• task-executor sometimes requested approval despite being in autonomous execution
  mode

Changes

Japanese Agents (.claude/agents-ja/)

• technical-designer.md:

  • Added implementation-approach.md to initial mandatory tasks
  • Clarified Phase 1-4 execution requirement for strategy selection
  • Specified L1/L2/L3 verification levels usage

• work-planner.md:

  • Added implementation-approach.md to initial mandatory tasks
  • Updated task decomposition to follow integration points and L1/L2/L3 levels

• task-executor.md:

  • Replaced abstract "3-location update" with concrete Edit tool instructions
  • Added progressUpdated field to structured response
  • Changed execution principle to explicit checkbox update requirement

English Agents (.claude/agents-en/)

• Applied equivalent changes with appropriate English translations
• Maintained consistency in technical terminology (L1/L2/L3, vertical/horizontal
  slice)

Impact

• ✅ Ensures proper metacognitive strategy selection process
• ✅ Prevents progress tracking omissions
• ✅ Improves AI execution accuracy
• ✅ Reduces need for unnecessary approval requests in autonomous mode