load_path behaviour should not depend on the current namespace

[ablaom]

import OutlierDetectionPython, MLJBase

MLJBase.load_path(OutlierDetectionPython.COFDetector)
# "OutlierDetectionPython.COFDetector"

using OutlierDetectionPython
MLJBase.load_path(OutlierDetectionPython.COFDetector)
# "COFDetector"

The correct load path (the first one) is being written to the MLJ model registry, which will explain why we didn't hit this earlier. However, this bug cased me some grief debugging a separate issue.

It'l likely this issue persists across the OutlierDetection ecosystem.

[davnn]

Hi, this is fixed now in OutlierDetectionInterface 0.2.1. Publishing with OutlierDetection.jl is still open because of other maintenance issues.