migrated from quarto-dev/quarto-cli#11656

Currently, the following bibliography.json file works as intended in a document that cites @osmnx and uses the citeproc filter:

[
    {
        "id": "osmnx",
        "type": "article-journal",
        "title": "OSMnx: New methods for acquiring, constructing, analyzing, and visualizing complex street networks"
    }
]

However, if we convert the JSON to YAML, no references will be loaded from bibliography.yaml:

- id: osmnx
  type: article-journal
  title: 'OSMnx: New methods for acquiring, constructing, analyzing, and visualizing complex street networks'

The workaround is easy, to use a dictionary/object as the top-level of the YAML with a key for references whose values are the array of CSL data references:

references:
- id: osmnx
  type: article-journal
  title: 'OSMnx: New methods for acquiring, constructing, analyzing, and visualizing complex street networks'

Accepting an array of references via YAML would make sense because

• for parity with the JSON handling
• since the official CSL Data spec specifies an array
• it used to work in older versions of pandoc (if I'm not mistaken)

Noting extra context provided by @jgm at quarto-dev/quarto-cli#11656 (comment):

Not really intentional, nor unintentional. I reused a function yamlToRefs that is defined in the Markdown reader. This is what the Markdown reader uses to extract references from the metadata, and obviously it needs to look for a references key. This is turn is just a wrapper around yamlBsToRefs (which operates on a bytestring). It would probably be harmless to adjust yamlBsToRefs, which currently just returns an empty list of references if the YAML is not a dictionary/object, to also handle the case where we have an array.