docs: fix PR template's ownership section to match reality for agent-…

[mattekudacy]

…opened PRs

Every PR this Claude Code session has opened (#7-#10) carries the Co-Authored-By trailer and session-link footer the template's own "Ownership" section says not to add, with a manually-written note explaining the conflict each time. That's not honesty about a constraint, it's the template silently failing its stated purpose on every single PR from this session — the checkbox is unsatisfiable and gets left unchecked with a workaround note instead.

The template was written for one case: a human contributor using an AI assistant to help draft their own PR, which they then submit and own. It never accounted for the other case this repo now actually has: an automated agent session opening PRs directly, where the attribution is a platform-level requirement the session cannot suppress from inside a PR body no matter what the template asks.

Splits the "Ownership" section into the original paragraph (unchanged in substance, narrowed to explicitly name the human-contributor case it was written for) plus a new, explicit exception for PRs an automated agent session opens directly — naming why the footer can't be omitted, and being equally explicit about what the exception does NOT cover: every other checklist item, the actual diff, tests, and docs still apply in full, and a human still decides whether to merge. Updates the checklist's attribution item to point at the exception instead of contradicting it, and asks future PRs in this category to reference the clause rather than re-explaining the same tension in prose every time.

Claude-Session: https://claude.ai/code/session_01M4WNEkbnKSx9mTg5Q1jX39

Ownership. This PR is submitted and owned by a human contributor, not an AI tool. Using
an AI assistant to help write the change is fine — but you reviewed every line, you
understand why it's correct, and you're accountable for it the same as any other
contribution. Do not add Co-Authored-By: trailers, "Generated with [tool]" badges, or any
other AI-attribution footer to commit messages or this description. A footer like that
implies the tool shares authorship or accountability; it doesn't, and code review here treats
every PR as fully human-owned regardless of how it was drafted.

What does this change and why?

Related issue

---

Type of change

• Bug fix
• New feature (failure type, recovery strategy, classifier, adapter, checkpoint store, ...)
• Breaking change (touches anything covered by the stability commitment —
  FailureType members/values, Step/FailureContext fields, RecoveryAction/FailurePolicy
  constructor and kwarg names, Classifier.classify(), Agent.__init__, CheckpointStore, adapter signatures)
• Documentation
• Classifier / corpus change (triage/classifier/rules.py, triage/scorer/, tests/data/error_corpus_*.json)

---

Checklist

• No AI-attribution trailers or badges anywhere in this PR — commit messages and this
  description are clean. I am the author and I am responsible for this change.
• pytest tests/ -x --tb=short passes locally
• ruff check ., ruff format --check ., and mypy triage/ --strict are all clean
• No new imports of openai/anthropic/langchain/langgraph/opentelemetry/etc. inside
  triage/ core — only triage/adapters/ and triage/observability/ may import optional deps
• If this adds or changes public API: docs/api/ and CHANGELOG.md (under [Unreleased]) are updated
• If this adds a FailureType or RecoveryAction: followed the steps in
  CONTRIBUTING.md

If this touches rules.py or an error corpus — read
scripts/README.md's "Corpus discipline" first:

• I did not tune rules.py against the current held-out corpus's misses (check
  scripts/README.md for which corpus is currently frozen/held-out vs. training data)
• scripts/classifier_accuracy.py was re-run and any changed numbers are reflected in
  README / CLAUDE.md / docs/known-limitations.md — quote the held-out number together
  with its self-healing/routing-sensitive split, never the aggregate alone

---

Anything reviewers should look at closely?