The consensus displayed at the top of the alignment is a dynamic consensus so you are unable to assign it to have a fixed sequence. I recommend you instead try one of the following options
1) It feels wrong to me to be generating an alignment containing a mixture of original and optimized sequences. How about generating an alignment of only optimized sequences, but add annotations to each optimized nucleotide to indicate that it has changed. Use for example annotations of type SequenceAnnotation.TYPE_EDITING_HISTORY_REPLACEMENT.
2) If you really want a mixture of optimized and unoptimized sequences in the alignment, you could add the "Unoptimized Consensus" as yet another sequence in your alignment and tell the user to turn off the automatically generated consensus.
3) You could call SequenceDocument.setFiedValue(DocumentField.NUMBER_OF_MAPPED_LOCATIONS,2) on sequences that you don't want to contribute to the automatically generated consensus. The consensus options will then have a setting called 'Ignore reads mapped to multiple locations' which you can turn on.