Add prioritize_failure_hotspots workflow proposal action

[akgohain]

Motivation

• Provide a read-only, approval-gated workflow-agent proposal that surfaces proofreading/inference failure hotspots so reviewers can prioritize manual inspection and rework.
• Use simple, explainable heuristics over recent event history (no ML or state mutation) to keep recommendations transparent and safe.

Description

• Add server_api/workflow/proposals.py which implements PRIORITIZE_FAILURE_HOTSPOTS and propose_failure_hotspots(events, *, top_k=3) using heuristic scoring of failure-linked proofreading/inference events.
• Heuristics normalize event types, derive candidate keys from common ID fields or locations, and compute a weighted score from signals like fail/error/reject/undo, severity, requires_rework, and whether the event touched proofreading or inference flows.
• Return a proposal payload with proposal_type, requires_approval=True, mutates_state=False, action: "review_ranked_hotspots", and an explanation containing ranked candidates and human-readable reason strings.
• Export the API via server_api/workflow/__init__.py and add focused unit tests at tests/test_workflow_hotspot_proposal.py covering ranking, fallback behavior, and approval/mutation flags.

Testing

• Ran unit tests with PYTHONPATH=. pytest -q tests/test_workflow_hotspot_proposal.py which passed: 3 passed.
• Attempted uv run pytest -q tests/test_workflow_hotspot_proposal.py which fails in this environment due to editable package metadata for pytorch_connectomics (missing pyproject.toml/setup.py) and is unrelated to the new logic.
• Tests assert the proposal is approval-gated and non-mutating, returns ranked candidates with reasons, and provides a fallback recommendation when events are insufficient.

---

Codex Task

[akgohain]

Consolidated into draft PR #165 (feat/codex-wave-integration). Closing this source draft to keep review in one place.