Run manual CI instantiating components across real proposal data

[philsmt]

There is lots of data out there which may break all kind of assumptions we inevitably have to make when writing components. Similar to offline calibration where we have manual CIs to test against real detector data, we could have such a CI that tries to instantiate various general purpose components across a lot (all?) runs present on GPFS or dCache and the results, e.g. exceptions occuring.