workflow-fix: add stall_reason to _PR struct in backends/runpod.py

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a prose follow-up raised by the Claude code-reviewer during /issue #701 round 2 (tasks/completed/701/events.jsonl → epm:code-review v2, ts 2026-06-28T08:17:04Z, emitting agent: code-reviewer).

Goal

Restore stall_reason on the _PR struct so RunPodBackend.poll no longer raises AttributeError: '_PR' object has no attribute 'stall_reason' when constructing a PollResult at src/explore_persona_space/backends/runpod.py:389.

Workflow gap

• Bug observed: tests/test_router*.py / tests/test_backend_*.py produces 2 failing tests with AttributeError: '_PR' object has no attribute 'stall_reason' at src/explore_persona_space/backends/runpod.py:389 (stall_reason=raw.stall_reason). The line is identical on main; backends/ was never touched by #701. The Claude code-reviewer flagged this when running the full pytest suite during /issue #701 round-2 review.
• Why it is a workflow gap: the _PR (poll-result) struct is missing the stall_reason attribute the RunPodBackend.poll method assumes — a workflow-helper code-shape regression on the unified backend router (per .claude/rules/workflow-fix-on-bug.md § "Workflow surface" / "src/explore_persona_space/backends/*.py" listing). A planner has no way to satisfy the line without this fix.
• Confidence (emitter): medium (the symptom is clear, the fix is to add the attribute; whether stall_reason should be sourced from somewhere else or defaulted is the planner's call).

Proposed change (candidate diff sketch — refine in planning)

Add the missing stall_reason attribute to the _PR (poll-result raw) struct so the RunPodBackend.poll consumer at runpod.py:389 can read it. Two plausible patch shapes for the planner to choose between:

• (A) Add stall_reason: str | None = None as a dataclass field on _PR (the raw poll-result struct populated by the SSH/SQL probe) — same default as the PollResult field — and have poll_pipeline.py (or whatever populates _PR) thread the actual stall-reason value when known.
• (B) Default-source stall_reason from raw.__dict__.get("stall_reason") at the PollResult(stall_reason=...) construction site so a missing attribute degrades to None.

Option (A) is the cleaner structural fix (the field belongs on _PR); (B) is a one-line getattr(raw, "stall_reason", None) defensive patch.

Scope / surfaces

• Primary target: src/explore_persona_space/backends/runpod.py (the _PR struct definition + the RunPodBackend.poll construction site at line 389).
• Probably also the _PR populator (poll_pipeline.py or the SSH probe path that builds _PR) — the planner should grep for _PR( constructors and confirm whether the new attribute needs threading or just a default.

Constraints / invariants

• Workflow-surface only — never experiment code, configs/, or tasks/.
• tests/test_router*.py + tests/test_backend_*.py must PASS after the fix.
• scripts/workflow_lint.py PASS.
• If workflow.yaml or CLAUDE.md need to change to document the field's semantics, they stay consistent with the existing poll_pipeline.py stall-reason taxonomy (see SKILL.md Step 7 § "Missing failure_class — invoke the classifier script" + .claude/rules/failure_patterns.md).

Provenance

• workflow_fix_target: src/explore_persona_space/backends/runpod.py
• fingerprint: 9359cb308676
• related_task: #701 (the code-reviewer that surfaced this was reviewing #701; the bug itself predates #701 and is on main)

The Claude code-reviewer's full verbatim surfaced-prose:

One note on the test claim The implementer's "356 tests passed" was a targeted subset. The full suite shows 5 failed, 5391 passed. I verified all 5 are unrelated to this doc-only patch:

• 2 are a pre-existing main bug — AttributeError: '_PR' object has no attribute 'stall_reason' at src/explore_persona_space/backends/runpod.py:389 (line identical on main; backends/ never touched by the #701 branch).
• 3 are test-isolation artifacts (they pass when the 5 are run alone).
• None reference planner.md.

These do not block the merge but the orchestrator may want a separate kind: infra fix for the runpod.py stall_reason regression on main.