For calling small variants, the standard way is to simply call diploid genotypes. You can already do a variety of research with unphased genotypes. You may further phase genotypes with imputation, pedigree or with long reads/linked reads, but not many are doing this because phasing is more difficult, may add cost and may not always give you new insight into your data. For these analyses, we use a haploid genome. For human samples, the vast majority of "large-scale genomic studies" are done this way.
A diploid reference actually doesn't help much with reference-based analysis; it only complicates algorithms. What could help a lot is a population reference, which may be represented by a graph or a compressed full-text index or both. In theory, if you have a comprehensive population reference and a capable mapping algorithm, you may call extra variants that would not be callable with short reads. In practice, however, there are quite a few technical challenges. Handling population references is a research topic. There are no "standards" yet.
If the goal is to assemble a new reference genome from a diploid sample, we almost always prefer to produce a diploid assembly. Unfortunately, I believe there are no "standard" procedures, either. SuperNova from 10x genomics builds the diploid information into a graph. Falcon from PacBio uses "unzip". I don't think they have got widely used and evaluated so far.
PS: saw your edit while writing the above. The fact that the genome only represents one strand does not mean we have to create the complement strand explicitly in analyses. We do most of reverse complement on the fly in algorithms as well as in mind.
reference genomes are not truly haploid
That depends on how the reference is assembled. If you sequence a haploid sample (e.g. bacteria), your assembly will be haploid. If you sequence an inbreed lab strain that is almost homozygous (e.g. mouse and fruit fly), your assembly will be nearly haploid. If you sequence a diploid sample, your assembly is very likely to be a mosaic of the two haplotypes. In case of the human reference genome, it is more complicated. It is largely a mosaic of several humans by stitching ~150kb haplotypes from these samples.