Change metadata output structure for cv scores

[henrikny]

In stead of this structure:
{
'r2-score': {'fold-1': 0.8906508246565554,
'fold-2': 0.6522734270761165,
'fold-3': -0.20146812532018704,
'fold-max': 0.8906508246565554,
'fold-min': -0.20146812532018704,
'fold-std': 0.46885465917412633,
'fold-mean': 0.4471520421374949},

'r2-score-1903.R-29TT3018.MA_Y': {'fold-1': 0.9854236451093319,
'fold-2': 0.9813427997012671,
'fold-3': 0.9813659734248961,
'fold-max': 0.9854236451093319,
'fold-min': 0.9813427997012671,
'fold-std': 0.0019182902041100804,
'fold-mean': 0.9827108060784985},

'r2-score-1903.R-29TT3019.MA_Y': {'fold-1': 0.9897565192366287,
'fold-2': 0.9900011126953097,
'fold-3': 0.9890387669335431,
'fold-max': 0.9900011126953097,
'fold-min': 0.9890387669335431,
'fold-std': 0.00040839845654897373,
'fold-mean': 0.9895987996218271},
}

Could we do this structure?:
{'r2-score': {
'model':
{'fold-1': 0.8906508246565554,
'fold-2': 0.6522734270761165,
'fold-3': -0.20146812532018704,
'fold-max': 0.8906508246565554,
'fold-min': -0.20146812532018704,
'fold-std': 0.46885465917412633,
'fold-mean': 0.4471520421374949},

'1903.R-29TT3018.MA_Y':
{'fold-1': 0.9854236451093319,
'fold-2': 0.9813427997012671,
'fold-3': 0.9813659734248961,
'fold-max': 0.9854236451093319,
'fold-min': 0.9813427997012671,
'fold-std': 0.0019182902041100804,
'fold-mean': 0.9827108060784985},

'1903.R-29TT3019.MA_Y':
{'fold-1': 0.9897565192366287,
'fold-2': 0.9900011126953097,
'fold-3': 0.9890387669335431,
'fold-max': 0.9900011126953097,
'fold-min': 0.9890387669335431,
'fold-std': 0.00040839845654897373,
'fold-mean': 0.9895987996218271},
}}

I am not sure of the inflictions this will have on other parts of the framework.

The related code can be found here (I think):

gordo/gordo/cli/cli.py

Line 232 in e73b3ec

all_scores.append(f"{metric_name}_{score_name}={score_val}")