fix: revert incorrect base_url fallback logic that breaks task execution

PR xorbitsai#138 (cee1e49) introduced a provider compatibility system that
incorrectly overrode user-configured base URLs with default values.

The problematic code used:
    base_url=model.base_url or default_base_url_for_provider(...)

This fails when base_url is an empty string ("") because Python's `or`
operator treats empty strings as falsy, causing the fallback to activate
even when the user explicitly configured a base URL.

For OpenAI-compatible providers like DashScope that store base_url=""
in the database, this caused requests to be sent to api.openai.com
instead of their configured endpoints. As a result, task execution
failed to start completely - tasks showed blank responses with no
backend execution activity.

This fix reverts to direct provider string comparison and removes the
fallback logic, trusting the database configuration as-is.

Fixes task execution regression introduced in cee1e49 (PR xorbitsai#138)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: tanbro

src/xagent/core/model/chat/basic/adapter.py

@@ -1,10 +1,6 @@
 import os
 
 from ....model import ChatModelConfig, ModelConfig
-from ....model.providers import (
-    default_base_url_for_provider,
-    provider_compatibility_for_provider,
-)
 from ....retry import create_retry_wrapper
 from ..error import retry_on
 from .azure_openai import AzureOpenAILLM
@@ -23,25 +19,11 @@ def create_base_llm(model: ModelConfig) -> BaseLLM:
     if not isinstance(model, ChatModelConfig):
         raise TypeError(f"Invalid model type: {type(model).__name__}")
 
-    compatibility = provider_compatibility_for_provider(model.model_provider)
-
-    if compatibility == "openai_compatible":
+    if model.model_provider == "openai":
         llm: BaseLLM = OpenAILLM(
             model_name=model.model_name,
             api_key=model.api_key,
-            base_url=model.base_url
-            or default_base_url_for_provider(model.model_provider),
-            default_temperature=model.default_temperature,
-            default_max_tokens=model.default_max_tokens,
-            timeout=model.timeout,
-            abilities=model.abilities,
-        )
-    elif compatibility == "claude_compatible":
-        llm = ClaudeLLM(
-            model_name=model.model_name,
-            api_key=model.api_key,
-            base_url=model.base_url
-            or default_base_url_for_provider(model.model_provider),
+            base_url=model.base_url,
             default_temperature=model.default_temperature,
             default_max_tokens=model.default_max_tokens,
             timeout=model.timeout,
@@ -78,6 +60,16 @@ def create_base_llm(model: ModelConfig) -> BaseLLM:
             timeout=model.timeout,
             abilities=model.abilities,
         )
+    elif model.model_provider == "claude":
+        llm = ClaudeLLM(
+            model_name=model.model_name,
+            api_key=model.api_key,
+            base_url=model.base_url,
+            default_temperature=model.default_temperature,
+            default_max_tokens=model.default_max_tokens,
+            timeout=model.timeout,
+            abilities=model.abilities,
+        )
     elif model.model_provider == "xinference":
         llm = XinferenceLLM(
             model_name=model.model_name,

src/xagent/core/model/chat/langchain.py

@@ -10,10 +10,6 @@
 from langchain_openai import AzureChatOpenAI, ChatOpenAI
 
 from ...model import ChatModelConfig, ModelConfig
-from ...model.providers import (
-    default_base_url_for_provider,
-    provider_compatibility_for_provider,
-)
 from ...retry import ExponentialBackoff, RetryStrategy, create_retry_wrapper
 from .error import retry_on
 
@@ -101,16 +97,14 @@ def create_base_chat_model(
         raise TypeError(f"Unsupported Chat model type: {type(model).__name__}")
 
     temp = temperature if temperature is not None else model.default_temperature
-    compatibility = provider_compatibility_for_provider(model.model_provider)
 
-    if compatibility == "openai_compatible":
+    if model.model_provider == "openai":
         return ChatOpenAI(
             model=model.model_name,
             temperature=temp,
             max_tokens=model.default_max_tokens,
             api_key=model.api_key,
-            base_url=model.base_url
-            or default_base_url_for_provider(model.model_provider),
+            base_url=model.base_url,
             timeout=model.timeout,
         )
     elif model.model_provider == "zhipu":