My take on this discussion, coming from a research lab: Metadata isn't meta.
For example, in recordings of, say, blood pressure over time, it's
common to think about things such as participant identifiers,
acquisition dates, event markers, and sampling rates as "metadata,"
and the actual measurements as "data."
But really: those meta things aren't ancillary to data analysis;
they're essential in keeping analyses organized, and often important
parameters in running an analysis at all.
Breaking things down into data versus metadata I think, encourages a
false (and not very interesting) dichotomy. If information has a use,
call it what it is: data. Store everything that's useful.
If you don't yet have a use in mind for your data, then you have a
place to start working :)