troubleshoot: ensure cache detail files are always emitted, add --trace all auto-discovery, and expand interactive/usage docs

[mouxinqq]

Motivation

• Make the Router troubleshooting skill produce stable, predictable output files and links for Cache diagnostics even when there are no samples, to avoid broken links and missing detail files.
• Provide a convenient --trace all mode to auto-discover trace targets for full-trace workflows.
• Improve user interaction/usage docs and prompts to clarify time-range selection, trace input behavior, and output layout.

Description

• Updated documentation in .claude/skills/troubleshoot/SKILL.md to move important notes into the header and to expand interactive prompts for log path, time-range selection (including discrete AskUserQuestion options), trace input guidance, and default output directories.
• Enhanced cache report templates in references/report_templates.md to require section presence even when counts are zero and to add eviction/交叉诊断 requirements.
• Modified cache analyzer in scripts/analyzers/cache.py to always include summary sections and stable detail links, to emit human-readable placeholder messages when data is missing, and to write more granular detail sections (cache_session_stickiness.md, cache_suboptimal.md, cache_eviction.md, cache_fallback.md, cache_cross.md).
• Adjusted report assembly in scripts/troubleshoot.py so detail files for cache are constructed consistently (with placeholders when empty), and changed file-write guards to write these detail files when keys are present (checking is not None).
• Added --trace all support by updating CLI help text and implementing _discover_full_trace_targets in scripts/analyzers/trace.py to discover session/trace/request IDs and return a human-readable auto-discovery summary appended to the trace summary.
• Minor CLI/help and comment tweaks to reflect the new trace behavior and to document that --trace supports all.

Testing

• Ran automated smoke test: executed python3 .claude/skills/troubleshoot/scripts/troubleshoot.py <sample_log> --cache against a sample log with no cache samples and verified the report summary/troubleshoot_report.md and detail files (detail/cache_session_stickiness.md, detail/cache_suboptimal.md, detail/cache_eviction.md, detail/cache_fallback.md, detail/cache_cross.md) were produced and contained the placeholder messages; test succeeded.
• Ran automated trace discovery smoke test: executed python3 .claude/skills/troubleshoot/scripts/troubleshoot.py <sample_log> --trace all and verified that _discover_full_trace_targets returned targets and the main summary included the auto-discovery summary; test succeeded.
• Performed a basic end-to-end run combining --tail and --cache to validate time-window handling and file outputs; run succeeded.

---

Codex Task