Adjudicator: handle SP shells, and spot-check the sets that agreed

The spot-check of basis sets where ECCE and BSE already agree found a bug
in the harness rather than in any basis set.  6-311G came out at -331
Hartree for an oxygen atom, from BOTH sources -- an impossible number, and
identical from both, which is the tell that the tool was at fault.

Cause: a shell label says which angular momentum each coefficient column
carries.  "SS" is two s functions sharing primitives; "SP" is one s and
one p.  Taking shell[0] for every column turned every Pople SP shell into
two s functions, so the atom was computed in an all-s basis.  Fixed, and
6-311G now reproduces both built-ins exactly from both sources.

All six earlier adjudications were re-run against the fixed harness and
are unchanged -- none of those basis sets uses SP shells -- so the finding
that ECCE's Ahlrichs VDZ, Ahlrichs TZV and cc-pVQZ-DK data is wrong for
certain elements stands.

The spot-check itself then came out reassuring.  Against the codes'
built-ins, for oxygen:

    STO-3G      ECCE 0.0          BSE 2.8e-08
    cc-pVQZ     ECCE 4.7e-11      BSE 4.7e-11
    def2-TZVP   ECCE 0.0          BSE 0.0
    cc-pV5Z-DK  ECCE 1.1e-08      BSE 1.1e-08
    cc-pVTZ     ECCE 4.1e-11      BSE 4.1e-11
    cc-pVDZ     ECCE 0.0          BSE 1.1e-09
    6-311G      ECCE 0.0          BSE 0.0

So the shipped library is sound for ordinary use, and the errors found are
localised to particular elements of particular less-common basis sets
rather than being systemic.  That is a materially different situation from
what the raw ECCE-versus-BSE diff suggested, and it only became visible by
bringing in an independent reference.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EFMUNbw2UpRPmgsqviAowb

Author: andy

tools/basissets/adjudicate.py

@@ -57,10 +57,17 @@ def orca_newgto(element, shells):
     """
     out = ["NewGTO %s" % element]
     for shell, nprim, ncoef, rows in shells:
-        letter = shell[0]
-        if letter not in LETTER:
-            return None                       # SP and friends: not handled
+        #  Each coefficient column is its own contracted function, and the
+        #  shell label says which angular momentum each column carries:
+        #  "SS" is two s functions sharing primitives, "SP" is one s and
+        #  one p.  Taking shell[0] for every column turns an SP shell into
+        #  two s functions -- which silently produced an oxygen atom at
+        #  -331 Hartree instead of -74.8 before this was fixed.
+        letters = list(shell) if len(shell) == ncoef else [shell[0]] * ncoef
         for column in range(ncoef):
+            letter = letters[column] if column < len(letters) else shell[0]
+            if letter not in LETTER:
+                return None
             kept = [(r[0], r[column + 1]) for r in rows
                     if len(r) > column + 1 and float(r[column + 1]) != 0.0]
             if not kept: