fix(logging): move module-level initialization to explicit setup calls (#3874)

- Moved environment variable parsing and `setup_logging()` call from
module level to proper initialization points
- Added explicit `setup_logging()` calls in `server.py::create_app()`
and `library_client.py::AsyncLlamaStackAsLibraryClient.__init__()`

Module-level side effects are bad practice and can cause issues with
import order, testing, and circular dependencies. The previous
implementation ran logging setup on every import of the log module,
which is unpredictable and difficult to control.

---------

Co-authored-by: Claude <noreply@anthropic.com>

llama_stack/core/library_client.py

@@ -47,7 +47,7 @@ from llama_stack.core.stack import (
 from llama_stack.core.utils.config import redact_sensitive_fields
 from llama_stack.core.utils.context import preserve_contexts_async_generator
 from llama_stack.core.utils.exec import in_notebook
-from llama_stack.log import get_logger
+from llama_stack.log import get_logger, setup_logging
 from llama_stack.providers.utils.telemetry.tracing import CURRENT_TRACE_CONTEXT, end_trace, setup_logger, start_trace
 from llama_stack.strong_typing.inspection import is_unwrapped_body_param
 
@@ -200,6 +200,9 @@ class AsyncLlamaStackAsLibraryClient(AsyncLlamaStackClient):
         skip_logger_removal: bool = False,
     ):
         super().__init__()
+        # Initialize logging from environment variables first
+        setup_logging()
+
         # when using the library client, we should not log to console since many
         # of our logs are intended for server-side usage
         if sinks_from_env := os.environ.get("TELEMETRY_SINKS", None):