Add optional cluster_mode to run FastTag on clusters instead of individual cells

[valleyofdawn]

FastTag currently performs annotation at the single-cell level. For large datasets or early-stage exploration, it would be useful to support an optional cluster-based mode.

This feature would:

1. Compute marker-based scores on cluster averages rather than individual cells.
2. Assign labels to clusters and propagate to all member cells.
3. Greatly improve performance on large datasets.
4. Increase interpretability by labeling major cell groups.

Proposed interface:

python
Copy
Edit
fasttag(
adata,
marker_dict,
cluster_mode=True,
cluster_key='leiden', # or any other .obs key
...
)
Considerations:

1. Scores will be based on mean expression within each cluster.
2. Smoothing or voting steps may need to be disabled or adapted in this mode.
3. Include safeguards/warnings for low-resolution or mixed clusters.
4. This should be optional, preserving default per-cell behavior.