Take the ProtOr element from the CCD, not from the atom name

[dylanpulver]

_calculate_protor_radii() takes an atom's element from the first character of its atom name. Wrong when the element has two letters, or a hydrogen's name does not start with H.

On main @ af1c4bb:

vdw_radius_protor("MSE", "SE") -> 1.77  # sulfur's radius, for selenium (same for SEC)
vdw_radius_protor("PHG", "C4") -> 1.76  # mercury named 'HG' read as a bonded H
vdw_radius_protor("GLY", "CA") -> 1.88  # control, unaffected

Across the CCD, 262 atoms in 175 components change: 187 had their own element misread (139 Se, 23 H, rest Cu/Ni/Ca/Cd/Co/Cr/Na/Cl read as C or N); 75 are heavy atoms with a wrong hydrogen count, because a neighbour named HG/HF counted as hydrogen or one named 1HZ/D1 did not. sasa(vdw_radii="ProtOr"), the default, calls this, using its 1.8 fallback where None is now returned.

The element now comes from chem_comp_atom.type_symbol, the same CCD entry the bonds come from. vdw_radius_single() already uses the annotated element; only this path re-derived it.

test_protor_radii cannot see it: it runs 1GYA filtered to standard amino acids — all one-letter elements — and only asserts the radius is not None.

tests/structure collects 4197: 4184 passed / 13 skipped; reverting only radii.py fails exactly the 4 new cases. A naive "two-letter prefix if it is an element" fix instead breaks the existing test_protor_radii, as CA reads as calcium.

Drafted with Claude Opus 5 (claude-opus-5); every measurement was run.

[padix-key]

Thanks for fixing this oversight 👍. I left a few comments below, otherwise this PR looks good to me!

[padix-key]

I think this is the more idiomatic (and likely faster) way to convert a numpy scalar into a native Python value.

Suggested change

	str(atom_name): str(element).upper()
	atom_name.item(): element.item().upper()

[codspeed]

Merging this PR will not alter performance

✅ 109 untouched benchmarks
⏩ 14 skipped benchmarks¹

---

_{Comparing dylanpulver:fix-protor-element-from-ccd (fddf7b5) with main (bcaf9e9)}

Footnotes

1. 14 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩