fix: address review feedback on PR-TRACE-1a (#244)

Must-fix issues:
- parent_edge_id now stores edge IDs (format: from_id:to_id:edge_type:hop)
  instead of node IDs, enabling proper tree reconstruction
- Budget counting now only counts newly discovered nodes; moved increment
  after visited check

Suggestions addressed:
- test_trace_direction_required: now validates pydantic ValidationError
- test_trace_inbound_callers_depth_2: removed vacuous >= 0 assertion
- test_trace_parent_edge_id_chain: now verifies parent_edge_id format
  and validates it references an edge that reaches the current node
- Tautological hop check: simplified to assert 0 in hops and hops <= {0, 1}
- Removed hardcoded row_kind from _edge_attrs_for_row
- Documented include_unresolved as 1a no-op (like prune_roles)

Tests: 22 passed, 1 skipped

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

Author: HumanBean17