ANGLe deduplication — CNS toppar .param scan

Per-file scan of src/haddock/cns/toppar/*.param for duplicate ANGLe
statements: the same atom-type triple (under the reversal symmetry
ANGLe A-B-C ≡ ANGLe C-B-A, middle atom fixed) defined on more than one line.

Companion to BOND-DUPLICATES.md and the conflict report in
FIXME-param.md.

Method

For each line: strip {...} blocks and trailing ! comments, locate the
ANGL* keyword token (case-insensitive — ANGLe, ANGLE, angle, ANGL
all match; also inside the eval (...) ANGLe ... form), take the three following
tokens as the atom-type triple and the rest as parameters. Canonicalize the
triple by reversal so A-B-C and C-B-A collapse to one key, group by key, and
report every key seen more than once. Detector self-tested against planted
reverse-order, lowercase-keyword, and eval-prefixed duplicates.

Each group is tagged IDENTICAL (all occurrences share the same parameter
values — safe to dedup, no behaviour change) or CONFLICT (occurrences
disagree in force constant and/or equilibrium angle — last-wins in CNS; needs a
force-field decision, do not blindly delete).

Summary

9 of 11 files are clean. Two need attention.

1. protein-allhdg5-4.param — 1 identical duplicate (safe)

 537: ANGLe  CR1E CR1E CY2      500.00 {sd=     0.031}    119.6224
 539: ANGLe  CR1E CR1E CY2      500.00 {sd=     0.031}    119.6224   <-- exact duplicate of L537

Line 538 in between (CR1E CR1W CY2) is a different triple, so this is an
isolated copy-paste artifact. Both entries are byte-for-byte identical, so
removing line 539 changes nothing physically — a clean, safe dedup.

2. protein-CG-Martini-2-2.param — 1470 duplicated triples

All 1470 are reverse-order pairs (A-B-C vs C-B-A); none are same-order.
Breakdown:

• 420 IDENTICAL — reverse pairs carrying the same values; safe to collapse to
  one entry with no behaviour change. Example:

  1767: ANGLe Bda B0 B0   5.971  134
  1962: ANGLe B0  B0 Bda  5.971  134   <-- reverse of L1767, same values
  

• 1050 CONFLICT — reverse pairs with different values (this is exactly the
  set flagged in FIXME-param.md §2). Because an angle potential is geometrically
  symmetric and CNS matches angle parameters order-independently, both entries
  match the same physical bead triplet and last-wins silently resolves them.
  Example:

   787: ANGLe H0 B0 B0   167.184   96
  1957: ANGLe B0 B0 H0     5.971  134   <-- reverse, different force const AND angle
  

This is an auto-generated coarse-grained force field. As FIXME-param.md notes,
the likely cause is the generation pipeline emitting per-instance directional
angles that CNS then collapses — so the fix for the 1050 conflicts may belong
upstream in the generator, not in the .param file. Verify before editing.

Recommendation

1. protein-allhdg5-4.param L539 — remove the exact duplicate; trivially safe.
2. protein-CG-Martini-2-2.param — the 420 IDENTICAL reverse pairs can be
   collapsed mechanically. Of the 1050 CONFLICT pairs, the 294 within 2° can
   also be collapsed (first instance dropped, second kept — see the
   ANGL-conflicts-removed.md table below); the remaining 756 (≥ 4° apart)
   are a force-field decision and should be checked against how the file is
   generated first.

Update: this PR acts on that recommendation — the 421 identical duplicates
and the 294 within-2° conflicts have now been removed; the 756 larger conflicts
are left untouched. The scan figures above describe the original (pre-removal)
state.

Exact-duplicate ANGLe statements removed

The 421 identical-value duplicate ANGLe groups removed in PR #1652. Each row is one canonical triple (under reversal symmetry A-B-C ≡ C-B-A); Kept is the first occurrence retained, Removed the redundant duplicate. Both carry identical parameter values, so removal is behaviour-preserving.

protein-allhdg5-4.param — 1 exact duplicates

protein-CG-Martini-2-2.param — 420 exact duplicates

Total: 421 exact-duplicate groups removed.

Conflicting ANGLe pairs removed within 2° tolerance

The 294 reverse-order conflicting ANGLe groups in protein-CG-Martini-2-2.param whose equilibrium-angle difference is ≤ 2°. For each, the first (earlier) instance was removed and the second kept. CNS resolves repeated angle definitions by last-wins, so the second instance was already the one in effect — removal is behaviour-preserving. Line numbers are from the file state just before this removal.

Total: 294 conflicting pairs collapsed (first instance removed, second kept).