If 0 length is passed in, we should return math.NaN like python code
Also follow SEP (Somebody else's problem)

Basically ported over https://github.com/grafana/metrictank/blob/master/expr/seriesaggregators.go
```
median
def safeMedian(values):
  safeValues = [v for v in values if v is not None]
  if safeValues:
    sortedVals = sorted(safeValues)
    mid = len(sortedVals) // 2
    if len(sortedVals) % 2 == 0:
      return float(sortedVals[mid-1] + sortedVals[mid]) / 2
    else:
      return sortedVals[mid]

diff
def safeDiff(values):
  safeValues = [v for v in values if v is not None]
  if safeValues:
    values = list(map(lambda x: x*-1, safeValues[1:]))
    values.insert(0, safeValues[0])
    return sum(values)

stddev
def safeStdDev(a):
  sm = safeSum(a)
  ln = safeLen(a)
  avg = safeDiv(sm,ln)
  if avg is None: return None
  sum = 0
  safeValues = [v for v in a if v is not None]
  for val in safeValues:
     sum = sum + (val - avg) * (val - avg)
  return math.sqrt(sum/ln)

range
  'range': lambda row: safeSubtract(safeMax(row), safeMin(row)),

multiply
  'multiply': lambda row: safeMul(*row),
def safeMul(*factors):
  if None in factors:
    return None

  factors = [float(x) for x in factors]
  product = reduce(lambda x,y: x*y, factors)
  return product
```

Note that this fixes bug --> unable to get cnt aggfunc
Also ignores archive which seems to be different