fix(agent): iteration-limit summary strips display_kind/display_metadata

[Kyzcreig]

Kanban t_e593f247 (found by Argus on #942 review; pre-existing on main).

handle_max_iterations builds its own api_messages and never popped display_kind / display_metadata, which the main-loop builder strips (#764 contract: they must never reach a model). A DB-reloaded confab-notice row leaked both keys, including the bridge request_id and detector grammar label, into the summary request. Strict OpenAI-compatible gateways reject unknown message keys.

Fix: pop both keys in the summary loop, and apply the same NousResearch#88955 neutral payload the main loop gives legacy empty hidden assistant placeholders.

Tests (tests/agent/test_iteration_limit_summary_display_fields.py):

• real path: turn 1 persists a confab row, turn 2 reloads from SessionDB with max_iterations=1, provider asks for a tool call, budget exhausts, summary fires; asserts on captured wire bytes (stream + non-stream).
• hidden-placeholder heal on the summary path.
  All 3 red on main, green with the fix.

Overlap with #942: it adds 2 lines at the top of the same loop (is_metadata_only_tool_notice skip). Different hunk; no textual conflict expected.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[Kyzcreig]

🤖 merged-by: apollo · lane: kanban-merge-pass · gate: BYPASS: FleetReview paused by Ace 2026-09-22 (state/fleetreview-pause marker present) · why: t_e593f247: Iteration-limit summary request forwards display_kind/display_metadata (confab r; Argus off card review (Ace 13:08), CI green

[Kyzcreig]

🤖 merged-by: apollo · lane: kanban-merge-pass · gate: BYPASS: FleetReview paused by Ace 2026-09-22 (state/fleetreview-pause marker present) · why: t_e593f247: Iteration-limit summary request forwards display_kind/display_metadata (confab r; worker completed in place, CI green

[ang-prism]

FleetReview

Review: post-merge · head 9fb3b4c83a6c · duration 1m 34s
Profile: light (merit: default light: lines 204<800, files 2<1000000, hunks 2<1000000, no hot path) · policy: below-size-and-path-gates
Roster: B-assert-ctx → gpt-6-sol (openai), C-assert-xhigh → claude-code-opus-5-5 (anthropic), F → gpt-6-sol (openai), L6 → gpt-6-sol (openai)

Post-merge review (fleetreview:post-merge override): this reviewed the merge commit against its first parent — the bytes that already shipped. It is not a pre-merge gate pass.

profile: light (rule: default light: lines 204<800, files 2<1000000, hunks 2<1000000, no hot path) · round 0 · members: B-assert-ctx, L6, C-assert-xhigh, F · families: anthropic,openai

Confidence: 4/5

No issues found.

---

FleetReview provenance · models: B=gpt-6-sol, C=claude-code-opus-5-5, F=gpt-6-sol · cost: $0.82 · duration: 1m 30s · rounds: 1 · files examined: 2