fix: _ensure_hf_token swallows click.Abort, causing misleading warning

[alfredoclarifai]

Bug

When running clarifai model local-runner (or any command that calls _ensure_hf_token) from a directory without a config.yaml, the CLI produces a confusing sequence of outputs:

1. A correct [ERROR] log: config.yaml not found in model path.
2. A misleading [WARNING]: Unexpected error ensuring HF_TOKEN: (with an empty message)
3. A full AssertionError traceback from ModelBuilder._validate_folder

The expected behavior is that the CLI should abort cleanly at step 1 and never reach steps 2 or 3.

Root Cause

In clarifai/cli/model.py, the _ensure_hf_token function raises click.Abort() when config.yaml is missing (line 865), but the outer except Exception as e block (line 887) catches it:

try:
    ...
    if not os.path.isfile(config_path):
        logger.error("`config.yaml` not found in model path.")
        raise click.Abort()   # intended to stop execution
    ...
except Exception as e:
    # catches click.Abort too — swallows the abort and downgrades it to a warning
    logger.warning(f"Unexpected error ensuring HF_TOKEN: {e}")

Since click.Abort() has no message string, {e} renders as empty, producing the cryptic Unexpected error ensuring HF_TOKEN: warning. Execution then continues past _ensure_hf_token, and the next thing that fails is the ModelBuilder assertion — producing a noisy traceback.

Fix

Add a specific except click.Abort: raise before the general except Exception handler so the abort propagates as intended:

except click.Abort:
    raise
except Exception as e:
    logger.warning(f"Unexpected error ensuring HF_TOKEN: {e}")

Affected Commands

_ensure_hf_token is called by multiple CLI commands:

• clarifai model upload
• clarifai model download-checkpoints
• clarifai model local-runner
• clarifai model local-grpc
• clarifai model signatures