Derive chartmap rmajor from magnetic axis length

[krystophny]

Since #373 the Boozer chartmap reader derives rmajor from the file geometry as the (theta, zeta)-average of sqrt(x^2 + y^2) on the innermost rho surface. GVEC computes the equivalent reference radius differently: the length of the magnetic axis divided by 2*pi.

For a non-planar stellarator axis the two definitions differ. The axis of W7-X type configurations makes helical excursions, so its length exceeds 2*pi times the average cylindrical radius by a few percent. rmajor enters only the integrator step scale, dtaumin = 2*pi*rmajor*1e2/npoiper2, so the axis-length definition paces the timestep by the curve a passing particle actually follows, which is the more natural normalization of npoiper2.

Proposal:

• Approximate the axis from the innermost rho surface: center the surface per zeta plane by theta-averaging x, y, z, then measure the length of the resulting closed curve and divide by 2*pi.
• Apply the same definition in the chartmap reader and document it next to the existing derivation.
• Consequences stay within the tolerance contract from Add booz_xform to Boozer chartmap converter; derive rmajor from geometry #373: dtaumin and ntau shift by a few percent; the 5 percent dtaumin check against VMEC-direct runs still holds for typical configurations. Verify on the QA and W7-X test cases before switching.

Low priority; the current average-R derivation is correct for its purpose. Worth doing when the chartmap path sees heavy use in cross-code benchmarks (#370), where consistent step pacing across field sources reduces one more degree of freedom.