My reference includes the primary chromosomes (1-22, X, Y) plus chrM, all unplaced and random contigs + the EBV decoy. See also Which human reference genome should I use?. ALT contigs are typically not included unless you intend to use an ALT-aware alignment pipeline, such as the one in BWAkit. That is typically not necessary unless you are interested in exactly these regions and the variation going on there. Therefore, for standard purposes, do not include them into the reference. The reason is that they represent alternative sequences for regions that are already included in the primary assembly (highly variable regions like MHC) leading to multimapping events between the primary assembly and the ALTs. As most downstream tools typically exclude multimappers, including the ALTs in an ALT-unaware alignment pipeline, eventually leads to a loss of these reads. See also Reference Genome Components.

Once you have all the chromsomes and contigs you want to include as fasta, cat them together into one file and index them with BWA. See again Which human reference genome should I use?.