[Opt] Optimize PR 5086 at function findEnvironmentOfRadiusN

[IchiruTake]

According to OpenSmiles, I found that the bond/edge exploited the valence bond theory for the chemical graph, and the optimization I attempted to done here is to reduce the next looping of algorithm by assuming the hydrogen atom only have 1 single bond attached to it (It would fail if hydrogen can connect two or more atoms: Figure 3: https://en.wikipedia.org/wiki/Hydrogen_bond)

In the PR 5086, the change is to convert from (in line 595)

for (const auto bond : mol.atomBonds(mol.getAtomWithIdx(oAtom)))

into something as this:

if ((optMode && mol.getAtomWithIdx(oAtom)->getAtomicNum() != 1) || !optMode) {
    for (const auto bond : mol.atomBonds(mol.getAtomWithIdx(oAtom)))
          ...
} 

where optMode is the new introduced argument fed by user (default to False to prevent backward incompatibility) (I don't have any name for this argument currently). Description: optMode (bool): Optional argument that assume the hydrogen atom can only have one neighbor (one bond) only. By setting to True, the algorithm can have a visible speed-up if your molecule use AddHs() function but would fail if the previous hypothesis above is incorrect. Default to False.

Why I believe this can work:

• The RDKit algorithm normally deal with one molecule only (not poly-molecule) and applied valence bond theory (a common theory for most situations).
• The speed up is achieved if the AddHs() function is implemented. The effect of this condition: if (useHs || mol.getAtomWithIdx(bond->getOtherAtomIdx(oAtom)) in line 596 is ignored.
• I think this algorithm is used at table bitInfo applied in ECFP algorithm.
• I have tested on Python and this passed all unit tests currently (I have read the OpenSmiles document but unable to make a SMILES to cross-check my idea). I am not familar with mol block MDL as V2000 or V3000.

Thank you for your reading and response.
Reference: #5086

[IchiruTake]

I suppose the better name of this argument is assumeIsolatedHydro.
The test of this argument can be easily applied on even methane (CH4) by function FindAtomEnvironmentOfRadiusN()
This argument is equivalent of treating the hydrogen atom to be isolated (directed-edge connected).

If you want to achieve both speed-up but not losing the information when the argument assumeIsolatedHydro is not introduced, we can write some thing as this in the user-client (default is still assumeIsolatedHydro=False). The failed case I can possibly think of would be [H2] or anything molecular graph that the hydrogen atom has more than one connection.

def PyWrap_FindAtomEnvironmentOfRadiusN(mol, atomIdx, radius, useHs, *args, **kwargs):
     atom = mol.GetAtomWithIdx(atomIdx)
     if atom.GetAtomicNum() == 1:
          return FindAtomEnvironmentOfRadiusN(mol, atomIdx, radius, useHs, assumeIsolatedHydro=False, *args, **kwargs)
     return FindAtomEnvironmentOfRadiusN(mol, atomIdx, radius, useHs, assumeIsolatedHydro=True, *args, **kwargs)

Test:
FindAtomEnvironmentOfRadiusN(atom='C', radius=1, assumeIsolatedHydro=True) -> [0, 1, 2, 3]
FindAtomEnvironmentOfRadiusN(atom='C', radius=1, assumeIsolatedHydro=False) -> [0, 1, 2, 3]

FindAtomEnvironmentOfRadiusN(atom='H', radius=1, assumeIsolatedHydro=True) -> []
FindAtomEnvironmentOfRadiusN(atom='H', radius=2, assumeIsolatedHydro=True) -> []
FindAtomEnvironmentOfRadiusN(atom='H', radius=1, assumeIsolatedHydro=False) -> [0]
FindAtomEnvironmentOfRadiusN(atom='H', radius=2, assumeIsolatedHydro=False) -> [0, 1, 2, 3]