Five knot-splitting spellings duplicate two canonical entry points, and only the canonical ones are usable

[gsdali]

Does not gate the next release. Found in the #480 census, deliberately left out of PR #561 to keep
that fix to the continuity vocabulary.

What

GeomConvert_BSplineSurfaceKnotSplitting and Geom2dConvert_BSplineCurveKnotSplitting are each
wrapped twice, by two families added three releases apart, which nothing has ever reconciled:

OCCT class	canonical	duplicate (v0.105.0)
GeomConvert_BSplineSurfaceKnotSplitting	Surface.knotSplitting(uContinuity:vContinuity:) (counts + parameters, #403)	Surface.bsplineKnotSplitsU(continuity:), bsplineKnotSplitsV(continuity:), bsplineKnotSplitValues(continuity:)
Geom2dConvert_BSplineCurveKnotSplitting	Curve2D.splitIndicesAtDiscontinuities(continuity:)	Curve2D.bsplineKnotSplits(continuity:), bsplineKnotSplitValues(continuity:)

Five Swift methods over five bridge functions (OCCTBSplineSurfaceKnotSplitsU/...V/
OCCTBSplineSurfaceKnotSplitValues, OCCTBSplineCurve2dKnotSplits,
OCCTBSplineCurve2dKnotSplitValues), all reaching the same two analyzers as the canonical pair.

The duplicates are strictly weaker, not differently useful:

• The two bsplineKnotSplits* count methods return exactly uSplitCount/vSplitCount and the
  canonical array's .count.
• Curve2D.bsplineKnotSplitValues returns the identical indices splitIndicesAtDiscontinuities
  returns, as [Int32] and empty rather than nil. PR fix(#480): one continuity vocabulary for the whole knot-splitting family, and the order it could not spell #561 added a test asserting they agree at
  every continuity, so this one is measured, not assumed.
• Surface.bsplineKnotSplitValues is the only one carrying anything the canonical call does not:
  knot-table indices rather than parameters. Those resolve through bsplineUKnot(index:) /
  bsplineVKnot(index:), so nothing is unreachable without it, and the global function they exist
  to feed (GeomConvert::SplitBSplineSurface) is not wrapped.
• Surface.bsplineKnotSplitValues also constructs the analyzer three times per call (once per
  count method, once for the values), where the canonical call constructs it once.

Both duplicates take one continuity for both parametric directions where the canonical surface call
takes one per direction, so they cannot express a question the canonical one can.

Fix

• Deprecate the five methods with @available(*, deprecated, ...) pointing at the canonical
  spelling, implemented over it rather than over their own bridge functions
• Delete the five now-unused bridge functions (OCCTBridge is not an SPM product, so the C-layer
  break is free, per Surface/Curve3D/Curve2D batch-evaluation bridge functions: three unconsolidated implementation generations per type, and OCCTGridEvalSurfaceD0/D1 use the opposite U/V-major layout from OCCTSurfaceEvaluateGrid #486)
• Collapse the duplicated docs/reference/Document-Geometry-Constructors.md sections into a
  pointer at Surface-Advanced.md / Curve2D.md
• Decide whether Surface's index-returning form is worth keeping at all, given
  GeomConvert::SplitBSplineSurface is unwrapped and the parameters are directly usable

Refs #480, #403, #377.

[gsdali]

Fixed by #589 (merged into refactor/381-pass1b).

Five v0.105.0 knot-splitting spellings (Surface.bsplineKnotSplitsU/bsplineKnotSplitsV/bsplineKnotSplitValues, Curve2D.bsplineKnotSplits/bsplineKnotSplitValues) duplicated the canonical Surface.knotSplitting/Curve2D.splitIndicesAtDiscontinuities pair, added three releases apart. All five are now deprecated and forward to their canonical sibling; the five duplicate bridge functions are deleted.

The premise that the five were strictly weaker did not survive measurement: Curve2D.splitIndicesAtDiscontinuities had a 256-entry silent truncation bug (returned the written count, not the true count — indistinguishable from a curve with exactly 256 splits) that the deprecated spelling didn't share. Forwarding without fixing it would have regressed the deprecated pair, so OCCTCurve2DSplitAtDiscontinuities now reports the true count and retries at it (the #481 contract). The one thing the deleted family carried that the canonical calls didn't — raw knot-table indices — is preserved: KnotSplitResult now carries uSplitIndices/vSplitIndices alongside the parameters, from the single analyzer construction that was already happening.

Note (via #589's review): this closing comment doesn't auto-link on the project board since Closes only populates closingIssuesReferences for PRs based on the default branch — this PR targeted refactor/381-pass1b.