Split report outputs into summary/detail, add session_id fallback and markdown report improvements

[mouxinqq]

Motivation

• Prevent overwriting historical exports by making --output an output base and creating timestamped summary and detail subdirectories per run.
• Improve session-level analysis by falling back to trace_id when session_id is absent and surfacing the identifier type.
• Make exported reports easier to read and navigate by producing Markdown tables, summary/detail split, and intra-report links for session switch details.

Description

• Updated documentation in SKILL.md for stat-cache-hitrate and troubleshoot to document that --output is treated as a base directory and that runs create <timestamp>/summary and <timestamp>/detail subfolders.
• In session_analysis.py changed session aggregation to use session_id if present or fall back to trace_id, store a unified session key as identity and add id_type to rows.
• In stat_cache_hitrate.py added helpers _render_markdown_table, _truncate_text, and _seq_label, removed the coverage field from compute_session_hitrate output, and updated report rendering to write main report to summary/cache_hitrate_report.md and per-window/session details to detail/*, including a prioritized session table, id-to-session mapping, and per-session switch sections.
• In troubleshoot/scripts/troubleshoot.py changed save_detailed_report to write to summary/troubleshoot_report.md and to emit details under detail/, and updated main logic to assemble output_base + timestamp into the run output path.

Testing

• No automated tests were run for these changes.

---

Codex Task