New overlapping criteria, fluffiness, and prm cleanup

[fpavogt]

Description:

This PR introduces a series of updates to the code, following the suggestions by @loforest in #83, including:

• the default value for OKTA_LIM8 has been changed to 99 following the large scale tests of @loforest.
• the GROUPING affinity has been changed from manhattan to euclidean, since the former is detrimental to rising/dropping layers.
• the "vertical extrema" of slices (to assess if they overlap) are now computed as min/max +- alt_pad_perc * (thickness). alt_pad_perc is a new ampycloud GROUPING parameter that replaces overlap !
• the minimum-required separation between layers can now be adjusted as a function of altitude, and is no longer hardcoded to +- 10'000 ft.
• the cloud base is now computed as a percentile of the total amount of points (starting from the smallest ones), with the option to define a lookback percentile to only use a fraction of the most recent measurements.
• the metarize routine now also computes the "fluffiness" of layers (see Compute the fluffiness of cloud layers #89). This requires to add scipy and statsmodels to the list of required dependencies.
• the list of user-tunable parameters has been cleaned from all the superfluous ones.
• the parameter OKTA_LIM0 has been replaced by MAX_HITS_OKTA0, which now specifies the maximum (absolute) number of hits in a given slice/group/layer for it to be considered to be 0 oktas.
• the parameter OKTA_LIM8 has been replaced by MAX_HOLES_OKTA8, which now specifies the maximum (absolute) number of holes (= non-detection of cloud bases) in a given slice/group/layer for it to be considered to be 8 oktas.
• layers are now required to be 250ft apart (below 10'000 ft) in order for a group to be split.

It also contains other less critical changes, including:

• improved docstrings
• more robust tests
• bug fix in step_scale()

Error(s) fixed:

Fixes #85 (scientific validity tests are now always run fully, even if one fails). Fixes #84. Fixes #89. Fixes #90. Fixes #87. Fixes #92. Fixes #98.

With the merging of #101 this also: fixes #102, fixes #95, fixes #100, fixes #93, and fixes #99.

Checklists:

• New code includes dedicated tests.
• New code has been linted.
• New code follows the project's style.
• New code is compatible with the 3-Clause BSD license.
• CHANGELOG has been updated.
• AUTHORS has been updated.
• Copyright years in module docstrings have been updated.

[fpavogt]

Hey @loforest, @regDaniel, @pjuda, @sibalm - fyi, this PR is now ready once again to have its scientific stability/validity evaluated. I have just merged #101 into it (which included a handful of small "logistical" Actions/Tests tweaks) so that all the tests are now green again.

Happy to talk live about the stability tests when you want, and in particular the necessity (maybe?) to fine-tune some of the parameters (old and new).

Side-note: the pytest action will be triggered twice per week as soon as this gets merged into develop. Only then will its result be accessible via RestAPI.

[pjuda]

What would be a minimal test case with an output which is not NCD but actually some cloud layers identified?