fix: strip extra_content from tool_calls for strict APIs (Fireworks, Mistral)

Fireworks/Mistral reject HTTP 400 'Extra inputs are not permitted, field:
messages[N].tool_calls[M].extra_content' on any session whose history
contains prior Gemini tool calls. Gemini 3 thinking models attach
extra_content (thought_signature) to tool_calls; it survived to the wire
because the sanitize paths only stripped call_id/response_item_id.

Strip extra_content from the outgoing wire copy in both sanitize paths
(ChatCompletionsTransport.convert_messages + _sanitize_tool_calls_for_strict_api),
but gate it on the target model: keep extra_content for Gemini-family
targets (the thought_signature MUST be replayed or Gemini 400s), strip it
for everyone else — including non-Gemini models that inherit a stale Gemini
signature earlier in a mixed-provider session. Native Gemini is unaffected
(GeminiNativeClient bypasses these paths).

Original stored history is never mutated (only the per-call copy).

Fixes #17986.

run_agent.py

@@ -4627,7 +4627,7 @@ class AIAgent:
         return reapply_reasoning_echo_for_provider(self, api_messages)
 
     @staticmethod
-    def _sanitize_tool_calls_for_strict_api(api_msg: dict) -> dict:
+    def _sanitize_tool_calls_for_strict_api(api_msg: dict, model: "str | None" = None) -> dict:
         """Strip Codex Responses API fields from tool_calls for strict providers.
 
         Providers like Mistral, Fireworks, and other strict OpenAI-compatible APIs
@@ -4636,17 +4636,26 @@ class AIAgent:
         the internal message history — this method only modifies the outgoing
         API copy.
 
+        ``extra_content`` (Gemini thought_signature) is also stripped — strict
+        providers reject it with "Extra inputs are not permitted" — UNLESS the
+        outgoing ``model`` is itself Gemini-family, in which case it must be
+        replayed (Gemini 3 thinking models 400 without it). Defaults to
+        stripping when no model is supplied.
+
         Creates new tool_call dicts rather than mutating in-place, so the
         original messages list retains call_id/response_item_id for Codex
         Responses API compatibility (e.g. if the session falls back to a
         Codex provider later).
 
-        Fields stripped: call_id, response_item_id
+        Fields stripped: call_id, response_item_id, extra_content (model-gated)
         """
         tool_calls = api_msg.get("tool_calls")
         if not isinstance(tool_calls, list):
             return api_msg
+        from agent.transports.chat_completions import _model_consumes_thought_signature
         _STRIP_KEYS = {"call_id", "response_item_id"}
+        if not _model_consumes_thought_signature(model):
+            _STRIP_KEYS = _STRIP_KEYS | {"extra_content"}
         api_msg["tool_calls"] = [
             {k: v for k, v in tc.items() if k not in _STRIP_KEYS}
             if isinstance(tc, dict) else tc