troubleshoot: add --trace all discovery, make cache details always emitted, and move trace details to detail/trace/

[mouxinqq]

Motivation

• Provide a convenient "full-trace" mode and clearer UX for trace collection so operators can request all and get an auto-discovered set of IDs to trace.
• Ensure cache diagnostics produce a stable set of detail files and consistent links even when some categories have no samples to avoid broken links and missing sections.
• Organize trace outputs into a dedicated subdirectory and update docs/UI prompts to match the new behavior and file layout.

Description

• Added support for the all (and synonyms) value in trace input by introducing _discover_full_trace_targets which auto-discovers session_id, trace_id, and standalone request_id targets and returns a concise discovery summary; integrated into analyze_trace and appended to the returned summary.
• Updated trace formatting and saving so per-trace detail files are written under detail/trace/trace_<ID>.md and links in reports point to detail/trace/ (changes in format_trace_report and save_detailed_report).
• Reworked cache report generation (format_cache_report and format_full_report) to always emit the individual cache detail sections and files (cache_session_stickiness, cache_suboptimal, cache_eviction, cache_fallback, cache_cross) with placeholder text like "未发现/样本不足" when no samples are present, and updated top-level summary links to include these files.
• Updated documentation and skill prompts: SKILL.md and references/report_templates.md were modified to clarify pre-run instructions, AskUserQuestion flows for time-range and trace inputs, examples, required detail files, and default paths; troubleshoot.py CLI help was updated to mention --trace all and adjusted logic for trace handling and detail file writing.

Testing

• No automated tests were executed as part of this change.

---

Codex Task