Ensure _temp dir does not interfere with pixel SOM and consensus cluster assignment

[alex-l-kong]

What is the purpose of this PR?

If a run is restarted from the beginning, or if a _temp dir still exists due to a crash, this creates a strange limbo state where the pipeline:

1. Will attempt to create an already existing _temp folder if an overwrite is forced (which is often the recommended solution in case the run is completely unsalvageable)
2. Cannot identify what all the seen SOM clusters are if all FOVs have already been assigned (because it simply "passes through" the assignment step, where this normally happens)

How did you implement your changes

To address these issues:

1. If a _temp directory is seen on overwrite=True, delete it first
2. If the pixel_pysom object contains no SOM references in som_clusters_seen, invokes a new function, pixel_cluster_utils.identify_seen_pixel_clusters, to populate this. The function does extend support to meta clusters as well, though this case is not directly invoked in the pipeline.

[camisowers]

Looks good! Just need to fix the builds.