fix(agent): match tool results on call_id||id in pre-request repair (#58168)

[kshitijk4poor]

Summary

Valid tool results are no longer dropped as false orphans by the pre-request repair pass when the assistant tool_call uses the Codex Responses id != call_id format — closing the HTTP 400 that permanently wedged long-running sessions (#58168).

Root cause: repair_message_sequence Pass 1 built its known-id set from only tc.get("id"), then matched tool results by tool_call_id. In Codex Responses format an assistant tool_call carries both id (fc_...) and a distinct call_id (call_...), and a tool result's tool_call_id may be keyed on either depending on which builder produced it. A result keyed on call_id looked orphaned, so the pass dropped it — leaving the assistant tool_call unanswered and triggering, on strict providers (DeepSeek, Kimi): Messages with role 'tool' must be a response to a preceding message with 'tool_calls'. Sessions that persisted such a sequence re-sent the orphan every turn and could never recover.

Changes

• agent/agent_runtime_helpers.py:
  • repair_message_sequence Pass 1 now registers both id and call_id for every assistant tool_call, so a tool result matching either key is recognized — consistent with AIAgent._get_tool_call_id_static and the compressor's _sanitize_tool_pairs.
  • _sanitize_tool_call_arguments (corrupted-args repair) now uses the same call_id || id precedence for its existing-result scan and stub insertion — it had the identical mismatch and would insert a duplicate stub that itself became an orphan.
• tests/run_agent/test_message_sequence_repair.py: 3 regression tests for the codex id != call_id case (match on call_id; match on only call_id; match on id when both present).

Validation

Scenario	Before	After
assistant id=fc_1 call_id=call_1, tool tool_call_id=call_1	tool dropped → assistant orphaned → 400	pair preserved, repairs=0
assistant only call_id, tool matches call_id	tool dropped	preserved
assistant id+call_id, tool matches id	preserved	preserved
genuine orphan tool (no matching assistant)	dropped	dropped (unchanged)
baseline OpenAI id==tool_call_id	preserved	preserved (repairs=0)
corrupted-args + codex-format existing result	duplicate orphan stub inserted	real result found, marker prepended, no orphan

• Targeted suites: 277 passed / 0 failed (test_message_sequence_repair, test_tool_call_args_sanitizer, test_strict_api_validation, test_provider_parity, test_context_compressor).
• E2E verified with real imports against a temp HERMES_HOME.
• ruff check: clean.

Fixes #58168