The upsides of this mechanism is that it is easier to use
and lighter weight than defining new FeatureSet classes and
methods to generate them. The downsides are that the metadata
are tags, rather than proper key-value pairs, as far as trackr
is concerned. (A backend could be engineered such that it
interpreted tags of the form 'key:value'
as key-value
pairs, but this won't occur without extra work, and thus those
implied fields will not be queriable via the trackr api
specifically. The values will be included in the metadata
generally though, so non-field-specific queries will work.