mv read loop hardcodes 4 instead of nnode

[sbryngelson]

Bug

In `src/pre_process/m_start_up.fpp`, the loop reading IB mesh vertex data hardcodes `do r = 1, 4` and uses offset `(i-1)*4` instead of `do r = 1, nnode` and `(i-1)*nnode`. This assumes all mesh elements are quads (4 vertices).

File: `src/pre_process/m_start_up.fpp`, lines 479-482

Impact: When nnode ≠ 4 (e.g., triangular elements with nnode=3), the loop reads wrong data and writes to wrong array positions, corrupting the IB mesh.

Severity: High — IB mesh loading is wrong for non-quad elements.

[sbryngelson]

Fix: #1189