For instance gzip with the --rsyncable option: it helps the rsync process and similar but anything that doesn't care can still unpack the compressed file with any old gzip decompression routine. There is a cost in that the dictionary resets result in a small reduction in compression rates for most content, but you've not introduced a new format.
The bioinformatics people noted above don't use this despite using gzip (instead using a container format for gzip packed blocks) because the result isn't directly seekable even though in theory you could start decompression at any point that the dictionary was reset. You could add an external index file that lists “output byte X is the start of a new cycle, at input byte Y” which would solve that without changing the gzip format itself, which might be how I'd have done things, but there are then issues of keeping the content and index file together, and verifying you have the right index, as data is passed around various places.
The zip format resets compression at file boundaries (each input file is compressed separately) and it keeps an index of where each file starts in the compressed output, hence you can see at the file level and unpack individual files, but doesn't index within each file so you can't use this to seek around the compressed version of a single large file. Resetting compression at file boundaries is why zip is particularly inefficient when compressing many small files, one of the reasons .tar.gz is preferred for source code archives even with the --rsyncable option.
So if it were possible (and practical) with the zstd output format, an indexed version of what gzip --rsyncable does could be a good compromise for a standard format that is seekable where it needs to be but with minimal reduction in compression. Assuming of course that zstd becomes as ubiquitous as gzip, available by default in most places, otherwise not needing different tools to read the compressed files is a moot point and the reduced compression becomes just a cost not a compromise.