feat: 动态 block 索引分配、OAuth 自动启停联动与 Antigravity 签名注入增强

- Gemini/OpenAI normalizer 改为按实际出现顺序延迟分配 thinking/text block 索引
- OAuth 失效标记时自动停用 Key，清除/刷新成功时自动启用
- Antigravity thought signature 注入支持 tool-specific 签名优先，再回退 session 级签名
- Claude normalizer 对非字符串 thinking block 降级为 UnknownBlock
- Gemini normalizer 反序列化时为 Antigravity 目标补充 signature
- signature_cache _prune 增加超限驱逐
- model/provider 包 __init__ 改为惰性导入避免循环依赖
- build_antigravity_url 复制 query_params 防止修改调用方原始字典
- upstream_fetcher build_all_format_configs 兼容非 EndpointFetchConfig 对象
- Antigravity HTTP 状态判定扩展与 Gemini endpoint check 支持 v1internal
- 新增 thought signature 注入与 signature cache 驱逐测试

Closes #165

Co-authored-by: AAEE86 <ppk0227@hotmail.com>

tests/api/handlers/base/test_antigravity_v1internal.py

@@ -204,3 +204,63 @@ async def test_antigravity_forces_conversion_path_in_stream_with_prefetch() -> N
 
     assert mock_convert.call_count >= 1
     assert any(b"data: {}" in c for c in out)
+
+
+def test_wrap_v1internal_request_injects_thought_signature_from_tool_cache() -> None:
+    from src.services.provider.adapters.antigravity.signature_cache import signature_cache
+
+    signature_cache.clear()
+
+    tool_id = "toolu_123"
+    sig = "a" * 60
+    signature_cache.cache_tool_signature(tool_id, sig)
+
+    gemini_request = {
+        "model": "gemini-2.0-flash",
+        "contents": [
+            {"role": "user", "parts": [{"text": "hi"}]},
+            {"role": "model", "parts": [{"function_call": {"name": "do", "id": tool_id}}]},
+        ],
+    }
+
+    wrapped = wrap_v1internal_request(
+        gemini_request,
+        project_id="project-123",
+        model="gemini-2.0-flash",
+    )
+
+    model_turn = wrapped["request"]["contents"][1]
+    part = model_turn["parts"][0]
+    assert part["thoughtSignature"] == sig
+
+
+def test_wrap_v1internal_request_injects_session_signature_when_tool_cache_missing() -> None:
+    from src.services.provider.adapters.antigravity.signature_cache import signature_cache
+
+    signature_cache.clear()
+
+    session_id = "sid-123"
+    sig = "b" * 60
+    signature_cache.cache_session_signature(session_id, sig, message_count=2)
+
+    gemini_request = {
+        "model": "gemini-2.0-flash",
+        "sessionId": session_id,
+        "contents": [
+            {"role": "user", "parts": [{"text": "hi"}]},
+            {
+                "role": "model",
+                "parts": [{"functionCall": {"name": "do", "id": "toolu_missing"}}],
+            },
+        ],
+    }
+
+    wrapped = wrap_v1internal_request(
+        gemini_request,
+        project_id="project-123",
+        model="gemini-2.0-flash",
+    )
+
+    model_turn = wrapped["request"]["contents"][1]
+    part = model_turn["parts"][0]
+    assert part["thoughtSignature"] == sig

tests/services/antigravity/test_signature_cache.py

@@ -1,5 +1,7 @@
 from __future__ import annotations
 
+from typing import Any
+
 from src.services.provider.adapters.antigravity.constants import DUMMY_THOUGHT_SIGNATURE
 from src.services.provider.adapters.antigravity.signature_cache import ThinkingSignatureCache
 
@@ -51,6 +53,22 @@ def test_tool_signature_short_ignored() -> None:
     assert cache.get_tool_signature("toolu_123") is None
 
 
+def test_tool_signature_cache_enforces_limit(monkeypatch: Any) -> None:
+    import src.services.provider.adapters.antigravity.signature_cache as sc_mod
+
+    # Use a small limit to make eviction deterministic in tests.
+    monkeypatch.setattr(sc_mod, "_TOOL_CACHE_LIMIT", 3)
+    cache = sc_mod.ThinkingSignatureCache()
+    cache.cache_tool_signature("toolu_1", _SIG_A)
+    cache.cache_tool_signature("toolu_2", _SIG_B)
+    cache.cache_tool_signature("toolu_3", _SIG_C)
+    cache.cache_tool_signature("toolu_4", _SIG_D)
+
+    # Oldest entry should be evicted once the limit is exceeded.
+    assert cache.get_tool_signature("toolu_1") is None
+    assert cache.get_tool_signature("toolu_4") == _SIG_D
+
+
 # ===== Layer 2: Thinking Families =====