.saveBN(...) -> .loadBN(...) round trip fails due to rounding in .toFast()

If I have a network with narrow bins in a DiscretizedVariable, the pickle process will create a BN that can't be loaded with loadBN. Here's some sample code:

import pyagrum as gum
import numpy as np
import tempfile, os, traceback

print(gum.__version__) # 2.2.1

X = gum.DiscretizedVariable("X", "X", [0.0, 100.000001, 100.000002, 200.0])
Y = gum.LabelizedVariable("Y", "Y", 2)
Y.changeLabel(0, "False")
Y.changeLabel(1, "True")

print(X.ticks()) # (0.0, 100.000001, 100.000002, 200.0)
print(X.toFast()) # Precision lost: X[0,100,100,200]

bn = gum.BayesNet()
nx = bn.add(X)
ny = bn.add(Y)
bn.addArc("Y", "X")

bn.cpt("Y").fillWith([0.5, 0.5])
bn.cpt("X").fillWith(1.0)
bn.cpt("X").normalizeAsCPT()

tmp = os.path.join(tempfile.gettempdir(), "tofast_roundtrip.pkl")
gum.saveBN(bn, tmp)
print("saved to", tmp)

# Will error out since `X[0,100,100,200]` looses a cut:
bn2 = gum.loadBN(tmp)

Here's the output:

2.2.1
(0.0, 100.000001, 100.000002, 200.0)
X[0,100,100,200]
saved to /var/folders/fh/xxvl5chx10g7ht__mkgq05t40000gn/T/tofast_roundtrip.pkl
---------------------------------------------------------------------------
SizeError                                 Traceback (most recent call last)
/var/folders/fh/xxvl5chx10g7ht__mkgq05t40000gn/T/ipykernel_70042/2284630757.py in ?()


     25 gum.saveBN(bn, tmp)
     26 print("saved to", tmp)
     27 
     28 # Will error out since `X[0,100,100,200]` looses a cut:
---> 29 bn2 = gum.loadBN(tmp)

~/.virtualenvs/bayes-validation-hH82x9Th/lib/python3.11/site-packages/pyagrum/pyagrum.py in ?(filename, listeners, verbose, **opts)
  26381     warns = bn.loadUAI(filename, listeners)
  26382   elif extension == "PKL":
  26383     import pickle
  26384     with open(filename, "rb") as f:
> 26385       bn = pickle.load(f)
  26386   else:
  26387     raise InvalidArgument("extension " + filename.split('.')
  26388     [-1] + " unknown. Please use among " + availableBNExts())

~/.virtualenvs/bayes-validation-hH82x9Th/lib/python3.11/site-packages/pyagrum/pyagrum.py in ?(self, state)
  13285             for father in state['parents'][son]:
  13286                 self.addArc(father,son)
  13287         self.endTopologyTransformation()
  13288         for node in state['cpt']:
> 13289             self.cpt(node).fillWith(state['cpt'][node])
...
   7801 
   7802         return self
   7803 

SizeError: [pyAgrum] incorrect size: Sizes do not match : 4!=6

I believe this is because nodes are serialized using [self.variable(i).toFast() for i in self.nodes()], and the Fast format doesn't preserve full precision.