Improve quality and comprehensiveness of metadata (through researcher education or by repository staff)
Zenodo record as an example: https://zenodo.org/record/7531
There you will see:
1) the link to the related publication (via its DOI)
2) the description field which is indexed and searchable (not sure if this is what you meant by "discovery layer”)
4) Minted DOI
5) Browse to Datasets and search only there
6) Several machine readable formats, including the DataCite link
7) Submitters can re-edit metadata, and curator can enrich
8) Harvested through OAI-PMH by several aggregators
Other important things demonstrated but not mentioned in your list are
+) The funder information
+) The rights information
Both of which are in the machine readable part as well
Here is the full metadata record for a Dryad data package (as it happens the most downloaded package): http://datadryad.org/handle/10255/dryad.38181
1) All Dryad data packages provide a collection of data relating to a specific publication. There are links to the article and if you scroll down the recommendation to cite both the article and the data package if the data is reused.
2) There is metadata for the package and descriptions of each of the data items. See the full metadata at http://datadryad.org/handle/10255/dryad.38181?show=full
4) Data package DOI (e.g. here it is doi:10.5061/dryad.6p76c3pb - note the dryad suffix).
Dryad only includes 'data packages' but these contain a wide variety of 'data' types. The metadata (including DOI) is machine readable. The funder and rights information is important. Dryad has rights information with CC0 recommended, but I'm not sure funder information is widely provided by Dryad. Authors are encouraged to provide as much appropriate metadata as possible at submission, but there is also a curation process.