Refactor load analyzer: extract report formatter, improve token/select-release parsing and worker-type handling

[mouxinqq]

Motivation

• Improve accuracy and readability of the Load Analyzer by fixing token/select-release parsing bugs and separating report formatting from analysis logic.
• Make select/release matching more robust by excluding token-only releases from request lifecycle matching and by normalizing worker types (prefill/decode/mixed).
• Provide both a short summary and a detailed report file for load analysis to be saved alongside other diagnostic details.
• Update skill prompts in SKILL.md to make AskUserQuestion options explicit.

Description

• Added analyzers/load_report.py which contains format_load_report(result) that returns (summary_text, detail_text) and renders tables/sparklines using chart utilities.
• Updated analyzers/load.py to import format_load_report, removed the embedded formatter, fixed regexes for token events, corrected regex group indexing, added _normalize_worker_type, and adjusted internal token parsing and _grep_lines pattern for token releases.
• Enhanced log_parser.py token/select regexes and match_select_release logic to normalize worker types, treat token releases separately, only use counter releases for matching request lifecycles, and produce per_worker and type_summary with token select/release counts and inferred token selects for mixed workers.
• Updated troubleshoot.py to accept the (summary, detail) output from format_load_report, include the detailed load report in the details dict, and save it as detail/load_select_release.md when present.
• Minor documentation tweaks in SKILL.md to clarify discrete AskUserQuestion options and defaults.

Testing

• No automated tests were added or run as part of this change.

---

Codex Task