Non-uniform measurement weighting

[fpavogt]

At the moment, ampycloud treats all the measurements equally. However, some users may be interested to apply stronger weights to (e.g.) the most recent measurements, in a manner similar to alternative codes out there. This discussion is meant to:
1) see if there's a strong interest to do that in ampycloud, and
2) figure-out how to best to it (i.e. maximum flexibility while still easy to use).

Here's a few points to keep in mind:

• The current slicing and grouping algorithms are not sensitive to weights - so the identification of slices/groups would never be affected by unequal weighting.
• The layering algorithm can be affected by "integer" weights, i.e. doubling/tripling/... points before running the Gaussian Mixture Model ... but is this really warranted ?
• Weights are important when computing the parameters of slices/groups/layers. In particular, one could imagine using weights to derive a slice/group/layer base altitude AND/OR sky-coverage-fraction. In this case, weights would only come into play when computing medians and means, and could thus potentially be non-integers ?

Possible implementation:
Users specify up to two additional columns in the input DataFrame, weight_alt_base and weight_skycov. If the columns are not specified, they are set to 1's by default in .utils.utils.check_data_consistency() (that would rely on a OPT_DATA_COLS dict in.hardcoded.py.

We would then simply need to take those weight into accounts when deriving the alt_base and perc in .data.CeiloChunk.metarize().

This approach has the advantage of being transparent to users that do not care, while leaving complete freedom to other users to define whichever weights they like.

[fpavogt]

As of #86, it is possible to specific a BASE_LVL_LOOKBACK_PERC to use only a sub-set of the (most recent) ceilometer hits to derive cloud base heights.