feat: OAuth 导入导出、提供商筛选、Gemini 图像生成支持与流式处理增强

- OAuth: 支持通过 Refresh Token 导入账号（文件拖拽/粘贴），OAuth Key 可导出为 JSON
- OAuth: 所有 OAuth 端点添加 require_admin 鉴权
- 提供商管理: 新增状态/API格式/模型三级筛选，后端返回 global_model_ids
- Gemini: 新增图像生成模型适配（finalize_provider_request 钩子 + envelope 跳过不兼容字段）
- 流式处理: buffer 残留数据 flush 与 token 兜底估算
- 上游元数据: 提取 merge_upstream_metadata，配额耗尽模型保留与深度合并
- Antigravity 配额: 无 quotaInfo 时视为耗尽，移除 Other 兜底分组
- README: 新增升级备份与回滚指南

tests/api/handlers/base/test_stream_processor.py

@@ -1,3 +1,4 @@
+import json
 from typing import Any
 
 from src.api.handlers.base.response_parser import (
@@ -34,3 +35,46 @@ def test_process_line_strips_newlines_and_finalizes_event() -> None:
     processor._process_line(ctx, sse_parser, "\n")
 
     assert ctx.has_completion is True
+
+
+def test_process_line_updates_openai_usage_from_usage_only_chunk() -> None:
+    ctx = StreamContext(model="test-model", api_format="openai:chat")
+    ctx.provider_api_format = "openai:chat"
+    processor = StreamProcessor(request_id="test-request", default_parser=DummyParser())
+    sse_parser = SSEEventParser()
+
+    usage_chunk = {
+        "id": "chatcmpl_test",
+        "object": "chat.completion.chunk",
+        "choices": [],
+        "usage": {"prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15},
+    }
+
+    processor._process_line(ctx, sse_parser, f"data: {json.dumps(usage_chunk)}\n")
+    processor._process_line(ctx, sse_parser, "\n")
+
+    assert ctx.input_tokens == 10
+    assert ctx.output_tokens == 5
+
+
+def test_process_line_handles_openai_usage_chunk_followed_by_done_without_blank_line() -> None:
+    ctx = StreamContext(model="test-model", api_format="openai:chat")
+    ctx.provider_api_format = "openai:chat"
+    processor = StreamProcessor(request_id="test-request", default_parser=DummyParser())
+    sse_parser = SSEEventParser()
+
+    usage_chunk = {
+        "id": "chatcmpl_test",
+        "object": "chat.completion.chunk",
+        "choices": [],
+        "usage": {"prompt_tokens": 7, "completion_tokens": 3, "total_tokens": 10},
+    }
+
+    # Some SSE implementations may emit consecutive data lines without an empty separator.
+    processor._process_line(ctx, sse_parser, f"data: {json.dumps(usage_chunk)}\n")
+    processor._process_line(ctx, sse_parser, "data: [DONE]\n")
+    processor._process_line(ctx, sse_parser, "\n")
+
+    assert ctx.input_tokens == 7
+    assert ctx.output_tokens == 3
+    assert ctx.has_completion is True