Improve handling of missing epochs in epochs

[t-b]

From #2109 (comment):

Problem with missing epochs could be solved like this: If no wildcard patterns are used in the textual epoch specification for epochs() then we can keep non-existing epochs as empty ranges. e.g. for sweep 18 we would return a (2,2) array where the range for E2 is [NaN,NaN]. Then in data() we skip sweep data extraction if the current range evaluated is [NaN,NaN].

If wildcards are used for epoch names in epochs we can not do that because we do not know how many matches we get for each sweep. We also do not know how the matches order up, such that a "auto indexed" addition may hit on unrelated ranges.

and from #2109 (comment):

skipping over missing epochs - using the full stimset name is okay (rather than wildcards), especially if there were a way to pass the acquired stimulus set names over zeroMQ
Generate a warning when data is missing