Fix authorization header setup logic in text embeddings inference

[jeesonwang]

PR Description
Problem:
The authorization header setup in the TextEmbeddingsInference class had inconsistent handling of authentication tokens. When auth_token was callable, it wasn't being properly processed before setting the Authorization header.
Solution:
Refactored the authorization header setup logic to:
Properly handle callable authentication tokens by invoking them with the base URL parameter
Extract the actual token value (whether from callable or static source)
Consistently format the Authorization header with the "Bearer" prefix
Changes:
Modified the token processing logic in base.py to ensure correct token extraction
Added proper handling for both callable and static authentication tokens
Ensured consistent Bearer token formatting in the Authorization header
Impact:
This fix ensures reliable authentication for the TextEmbeddingsInference integration, preventing potential authentication failures when using callable token providers.