Multi-OCR alignment: add method=custom with user-defined prioritization

[bertsky]

Unfortunately OCR engines and models vary widely w.r.t. quality of their confidences (uncalibrated probability estimation). The aligner therefore has a hard time reaching good decisions. But sometimes users would still be able to formulate a priority rule – "if >80% use OCR1, else if >90% use OCR2, else use OCR3".

We should support that by adding method="custom" and a new parameter priority as a list (JSON array) of floats of confidence thresholds for the corresponding input fileGrps.