TrendAnalysis uses power_expected for both sensor- and clearsky-based workflows

[kandersolar]

Not necessarily a bug, but a perhaps unintuitive behavior: when passing in an external power_expected time series, it is used for both sensor-based and clearsky-based workflows. To my mind, part of the spirit behind the clear-sky workflow is that the normalization is tied to the modeled clear-sky irradiance, which is not the case when using a custom power_expected. I think if passing in an external power_expected time series, the sensor and clear-sky workflows become identical except for the csi_filter (not 100% sure about this though).

Some ideas:

• Separate power_expected_sensor and power_expected_clearsky inputs
• Keep the code the way it is now, but emit a warning if power_expected is used in the clear-sky workflow
• Do nothing, because the code is just doing what you told it to do

[martin-springer]

Let's implement the warning.