trace-to-skill v0.1.103

Adds usage-doctor attribution for OpenAI Codex token-burn demand around #14593 and related usage-drain issues.\n\nWhat changed:\n- New usage-doctor alias for usage-evidence\n- Usage receipt now includes confidence-ranked attribution buckets\n- Buckets cover quota-window accounting, rapid-drain repros, prompt-cache collapse, large cached-context replay, background polling, compaction loops, retry/tool loops, subagent fan-out, and idle/background drain\n- Each bucket includes signal count, line-linked evidence, and next evidence to collect\n- JSON schema and OpenAI OSS brief updated\n\nValidation:\n- npm run check\n- YAML workflow parse\n- git diff --check\n- npm pack --dry-run\n- synthetic usage-doctor attribution proof\n- live openai/codex issue-map proof for #14593