Fixed stereochemistry of mitsunobu and added corresponding tests

[FabianKruger]

Mitsunobu proceeds by backside attack, so the carbinol carbon inverts. The templates
left that atom unmarked, and RDKit's default is to carry the reactant's configuration
into the product so every stereodefined Mitsunobu product was the wrong enantiomer.

results = enumerator.enumerate_pair("Oc1ccccc1", "C[C@H](O)CC")
for result in results:
    print(f"{result.reaction_name}: {result.products}")
# before: mitsunobu: ['CC[C@H](C)Oc1ccccc1']    retention, wrong
# after:  mitsunobu: ['CC[C@@H](C)Oc1ccccc1']   inversion, correct

All 60 mitsunobu templates now mark atom map 4 as [#6@:4] on the reactant side and
[#6@@:4] on the product side. RDKit reads differing tags as invert relative to the
input, so one template still covers both input enantiomers and an undefined centre
stays undefined.

These are the only chirality tags in the 9,594 templates, so no other reaction is
affected, confirmed by re-running the full case-study enumeration: mitsunobu
stereochemistry changes, everything else is byte-identical, no products gained or lost.
Tests cover both enantiomers and the undefined case.