As you mentioned you should have a list of of background and a list of 'hits' for performing GO based enrichment analysis. For full set of annotated genes from Zebra fish you may use the GOA (Gene Ontology Annotation) for Zebra fish from GOA project website at EBI.

The GO consortium maintain a list derived from ZFIN, described here. From Ensembl, they can be obtained through the Ensembl Biomart linked at the top of the D.rerio index page. The GO terms are found under the attributes section. I'm not so familiar with NCBI, so I can't offer any help there.

Unless you already know that a particular source of annotation is most reliable, I would obtain the data from all of them and make a comparison. You are not really measuring over/under represented biological functions, you're measuring over/under represented annotation of biological functions, which we hope is similar, but often isn't.