Upgrade of data training

[HealthyPear]

Requirements

• fix Travis error finding config file
• Update documentation
• Update the rest of the pipeline enough to not crash
• Enable CI on Github Actions (Enable CI from GitHub actions #84)

Description

This is a big upgrade, targeting various parts of the pipeline up to DL2/a.

• upgrade to the API of ctapipe 0.9.1 (this improved calibration and allows us to close Comparison with CTA-MARS: Calibration #31 )
• the data format is not exactly the same yet (only nomenclature of variables up to DL1 - see Implement new DL1 data format from ctapipe #25 for future improvements)
• write_dl1 has become data_training (1/2 of issue Give appropriate names to the scripts that produce data level information #53 )
• DL1 and images output are merged in a single file; images are always saved - when requested - for all events and all files
• scale correction with the effective focal length has been applied (Closes Camera scale transformation using effective focal length #44 )
• Closes Move all config files under same directory #56 .

[kosack]

This probably doesn't even need to be a function, right? Nothing inside is computed or changes. Why not just make it a module-level variable called CTAMARS_RADII?

[HealthyPear]

Yes, you're right.
Would you prefer I do it here or I keep it a small fix for later?

[kosack]

It's a bit hard to separate the main changes from re-formatting, but I guess the biggest changes are just to support the ctapipe-0.9.1 container name changes. This can be considered a first step on the way to fully using the new features of ctapipe-0.9.1 and above, like ImageProcessor and DL1Writer, which can come at a later time (or with the standardized DL1/DL2 tools in ctapipe).

Otherwise, I don't see any big problems so far, but obviously this will be a bit superficial review, as can't easily test that the results are good.

[HealthyPear]

It's a bit hard to separate the main changes from re-formatting, but I guess the biggest changes are just to support the ctapipe-0.9.1 container name changes. This can be considered a first step on the way to fully using the new features of ctapipe-0.9.1 and above, like ImageProcessor and DL1Writer, which can come at a later time (or with the standardized DL1/DL2 tools in ctapipe).

Otherwise, I don't see any big problems so far, but obviously this will be a bit superficial review, as can't easily test that the results are good.

Yes, the upgrade to 0.9.1 was the main reason as it changed the API in some places.
The integration test for DL1 works though (it is part of the CI pipeline)

[HealthyPear]

From the point of view of physical performance, this PR updates the calibration and image extraction (from which the latest results come - the ones that got temporarily reverted during between ctapipe 0.9.1 and 0.10.1) plus it enables the "optical aberration" correction as it is done in CTAMARS.

From the data point of view, the variables related to image parameters are the same as in ctapipe (even though the format is still a single table) and the (optional) images file is merged with the rest so 1 run from data_training will produce 1 file (like ctapipe)

plus other minor changes