Do not record the identity as a rotation axis in PointGroupAnalyzer

[HiroYokoyama]

Summary

I found an issue in the symmetry analyzer: when a molecule whose only rotational symmetry is a single C2 axis (e.g., tetramethylhydrazine) is analyzed, sch_symbol reports it as D2, while get_symmetry_operations() returns only 2 operations. D2 is a group of order 4, so this is a self-contradiction.
I then used Claude Opus 5 to analyze the issue, and found that _check_rot_sym() registers a 360° rotation (the identity, which is valid about any axis) in rot_sym as a rotation axis, so the condition len(rot_sym) >= 2 in _proc_sym_top() routes the molecule to the dihedral branch.

Measured data:

C2 rotation residual about each principal axis (max atom displacement):
   axis [ 0.9866  0.122 -0.1085]   residual = 1.4231 A
   axis [-0.1375  0.2628 -0.955 ]  residual = 1.4232 A
   axis [-0.088   0.9571  0.2761]  residual = 0.0001 A
Best over 40000 random axes at least 10 deg away from that axis: 0.7332 A

                      before      after
tetramethylhydrazine  D2 / 2 ops  C2 / 2 ops
water C2v/4, methane Td/24, ammonia C3v/6, allene D2d/8,
benzene D6h/24, H2O2 C2/2, ethene D2h/8, HCl C*v/1, CO2 D*h/2   (unchanged)

This PR stops the loop in _check_rot_sym() at 2, and also removes the if len(self.rot_sym) > 0: guard in _proc_sym_top() so that the perpendicular C2 search always runs; that guard was only ever satisfied by the phantom entry, so fixing the loop alone would make these molecules report as C1 instead.

This PR also adds a regression test for tetramethylhydrazine in tests/symmetry/test_analyzer.py; the test fails on main with assert 'D2' == 'C2'.

This may be a breaking change:
sch_symbol changes for molecules whose unique axis of inertia carries no rotation of its own but which have one perpendicular C2 axis: they were reported as D2 and are now reported as C2. get_symmetry_operations() and get_pointgroup() are unaffected — they already returned the 2-operation group. Code that compared sch_symbol against the previous (incorrect) value must be updated; nothing else is affected.

Checklist

• Tests for the affected code pass locally: pytest tests/symmetry (92 passed / 5 skipped, before the patch 91/5)
• Lint passes: ruff check / ruff format --check passed

Breaking changes?

• This PR contains breaking changes (removals/renames, signature or
  behavior changes, changed defaults/output, dropped Python versions, ...).
  If so, prefix the title with [breaking] and describe the change and
  migration steps under ## Breaking Changes below.

Breaking Changes

[shyuep]

Automated PR review generated by Claude (posted by @shyuep's review bot).

Diagnosis and fix are correct. idx=1 in _check_rot_sym registered a 360° rotation (the identity) as a rotation axis about any axis, so len(rot_sym) >= 2 misrouted single-perpendicular-C2 molecules to _proc_dihedral (D2 symbol with only 2 ops — self-contradictory). Stopping the loop at 2 and unconditionally running _check_perpendicular_r2_axis is the right pair of changes; either alone would be wrong (C1 regression), as the PR notes.

Two points to verify before merge:

1. rot_sym can now be empty after _check_rot_sym where the phantom identity entry previously guaranteed at least one element. Please audit the other callers (_proc_asym_top, _proc_cyclic, spherical-top path) for any self.rot_sym[0] access or len(rot_sym) > 0 logic that implicitly relied on the identity entry. The passing test suite suggests this is fine, but an explicit check is warranted since the change is in a shared helper.
2. The body states this may change sch_symbol for affected molecules but the breaking-change box is unticked — per repo convention, tick it and prefix the title with [breaking].

Test is well-constructed (regression on Schoenflies symbol, operation count, and rot_sym order). CI is green. LGTM modulo point 1.