[shreyanshj10]

What does this PR do and why?

Removes the A23a and A26a detection loops from bondglycans.cns.

Patches are applied at the end of that file with reference=- bound to the $id1
residue and reference=+ bound to the $id2 residue. PRESidue A23, A26 and
A26S all take the sialic acid as their - reference: they modify -C2, delete
-O2/-HO2, and add dihedrals through -C1, -O6 and -O1A. Of those, O1A only
exists on SIA/SIB.

A23a and A26a select (resname GAL or resname NGA) and name C2 as $id1, so they
pass the acceptor as -, which is the wrong way round:

• A23a looks for name O3 on SIA/SIB, and neither residue has an O3 (their oxygens
  are O1A O1B O2 O4 O6 O7 O8 O9 O10), so its inner loop iterates over an empty
  selection and the loop can never fire.
• A26a finds SIA O6, but that is the ring oxygen, so the 2.0 A test against the
  acceptor's C2 is not met by a correctly built linkage. If it ever did fire it would
  ask for -O1A on a GAL or NGA.

The correctly oriented A23c and A26c loops take SIA/SIB as $id1 and are what makes
sialylation work today, so removing the two backwards ones is behaviour-neutral.

I have left the -O5 improper in A26S/A28S from the follow-up comment alone: as you
note there, renaming it to -O6 needs a matching IMPRoper C OAS OES CCE target derived
first, so it belongs in its own change.

One thing worth a look while reviewing: with A26a gone, PRESidue A26 is no longer
referenced from bondglycans.cns - A26c applies A26S, which is the same patch plus
the alpha-anomer improper. I left A26 in carbohydrate.top since it is part of the
topology library and can still be applied by hand, but if you would rather it went too I
am happy to add that here.

How was this tested?

Added test_sialylation_loops_select_the_sialic_acid_as_first_reference to
tests/test_module_topoaa.py. It parses bondglycans.cns, picks out every outer loop
that assigns a sialylation patch (A23, A26, A26S, A28S), and asserts that each one
selects resname SIA or resname SIB and name C2 as $id1, and that $id2 selects the
acceptor oxygen the patch bonds to (O3 for A23, O6 for A26/A26S, O8 for
A28S). It needs no CNS executable, so it sits with the unit tests.

I ran those assertions against bondglycans.cns as it stands on main and against the
version in this PR: on main they flag A23a and A26a and pass for A23c, A26c and
A28a; with the two loops removed all three remaining loops pass. The test also pins the
count at three, so it cannot pass by finding nothing.

AI assistance

An AI assistant helped with the write-up and with drafting the parser in the new test. The
chemistry is not taken from it: the reference direction comes from the PRESidue A23,
A26 and A26S bodies and the SIA/SIB/GAL/NGA atom lists in carbohydrate.top, which I
read and checked by hand, and the removal itself is the one #1711 asks for.

Checklist

• Tests cover the new and/or changed code
• Documentation updated if needed (also in the haddock3 user-manual
• CHANGELOG.md updated for user-facing changes

No user-facing behaviour changes, so nothing to document.

Related issues

Fixes #1711

Notes for reviewers

The diff is 72 deletions in bondglycans.cns plus the test and a changelog line; the
remaining A23c/A26c loops are untouched, so this should not collide with the A2G work
in #1710 beyond the changelog.