Keep rmaga inside rzl_array for lasym with ntor = 0

[CharlesCNorton]

Fixes #9.

For LASYM = T with NTOR = 0, read_indata leaves rcs = 0 because the cos(mu)sin(nv) basis type does not exist, and rmaga => rzl_array(1,:,0,rcs) in eqfor.f90 then points at type index 0 of a dimension declared 1:3*ntmax. rmaga is now associated only when rcs > 0 and otherwise points at a zero array of the same extent, which is the value the missing basis type contributes.

Checked with input.up_down_asymmetric_tokamak (VMEC++ test data) and a symmetric tokamak boundary run with LASYM = T: the wout files are unchanged bit for bit, and a build with -fcheck=all that used to stop at eqfor.f90:73 with Index '0' of dimension 4 of array 'rzl_array' outside of expected range (1:6) now terminates normally.