Change shape of dataframe of `extract_dataframe` in morph_stats (#874)

neuron features in morph_stats (#874)

The features in the 'neuron' entry of the config dict will not be run on each neurite_type, but on the whole neuron only once. This avoid duplicating features if they are the same on all neurite types, such as the ones in neuronfunc.py

The returned dataframe is now of the form:
```  
       | property  | axon                  | all                   | neuron
       | name      | feature_1 | feature_2 | feature_1 | feature_2 | feature_3
   -------------------------------------------------------------------------------------
    0  | simple    | 0.1       | 0.2       | 0.15      | 0.25      | 1000
```

Change the shape of the output DataFrame of `extract_dataframe` (#862)

💥 Change the shape of the output DataFrame of `extract_dataframe` (#862)

`extract_dataframe` used to return `N` rows for each neuron where `N` is the
number of neurite types in the input config.

## Context

Since `extract_dataframe` input is a list of neuron, it is more natural to
return exactly one row per neuron.

## Proposed change

Use a MultiIndex on the columns to hold the metrics for each neurite type
instead of having the neurite type appear in the index (axis=0).

## Example: 

Before:

|   | name    | neurite_type    | metric A | metric B |
|---|---------|-----------------|----------|----------|
| 0 | no-axon | axon            | val A    | val B    |
| 1 | no-axon | apical_dendrite | val A'   | val B'   |

After:


|   |         | axon                | apical_dendrite            |
|   | name    | metric A | metric B | metric A        | metric B |
|---|---------|----------|----------|-----------------|----------|
| 0 | no-axon | val A    | val B    | val A'          | val B'   |

## :sparkles: Additional modification

The returned dataframe is now sorted as the input list of neuron and support multiple neuron with the same name

In practice

- `imap` is now used instead of `imap_unordered`. Maintaining the sorting is worth the slight performance decrease.
-  the intermediate results are no longer stored in a `dict` but in a `list`. This enables having multiple neuron with the same name. This is for example useful to compare releases.

Drop the fst/ deprecated folder

Relax h5py constraint (#857)

and remove the deprecated fst/ files

Pin h5py < 3.0.0

Fix h5py < 3.0.0 (#855)

So that packages having as dependencies both bluepy and neurom (like
morph-validator), get h5py < 3.
Otherwise bluepy breaks

Fix newer pylint constraints (#851)

v1.5.3

Create publish-sdist.yml

add multiprocessing to morph_stat.extract_dataframe

Use multiprocessing in morphs_stats (#847)

Small refactoring of morphs_stats module for computational speedup, with the following changes:

- uses pool.imap_unordered in extract_dataframe in the loop overs neurons
- moved neurom.get outside mode loop to avoid duplicate computations

New function extract_dataframe that extract morphometrics as a dataframe

New function extract_dataframe that extract morphometrics as a datafr…

…ame (#841)


if the optional argument as_frame=True, otherwise the legacy return type (dict) is used

The dataframe looks like:

     name     neurite_type  ...  max_segment_midpoint_Y  max_segment_midpoint_Z
0  Neuron             axon  ...                0.000000               49.520306
1  Neuron  apical_dendrite  ...                0.000000               53.750948
4  simple             axon  ...               -2.000000                0.000000
5  simple  apical_dendrite  ...                     NaN                     NaN


And the 'neurite_type' key becomes optional for `extract_dataframe` and
`extract_stats`

Drop python 2 support

Drop python 2 support (#829)

NeuroM is now supported from python 3.5 to 3.8

morph_stat can now extract all features with the '-f' flag (#817)

v1.4.20

Bump version to v1.4.20

All checks should return list of points (#807)

v1.4.19

Bump version number (#813)