Preserve helical source during reconstruction replay

[krystophny]

Summary

• apply the helical source during local reconstruction replay
• restore reconstructed incoming rows after all local source assembly
• initialize unused Legendre padding before whole-array collisionality scaling
• test both boundary directions and two Laguerre blocks

Closes #143. Stacked after #148 and remains unmerged with the programme stack.

Reconstructed incoming rows remain authoritative. The drift-kinetic operator, collision model, Runge-Kutta stencil, eta/Laguerre layout, sparse solve, convergence criteria, qflux normalization, memory layout, ABI, and generated dispatch are unchanged. Valid Legendre entries are overwritten exactly as before; only unused padding is initialized.

Verification

Test fails on main

$ fo exec --no-build --cwd "$RUN" neo_2_ql.x
ERROR STOP mode-1 helical drive does not support reconstruction mode

The focused boundary-precedence fixture also fails to compile on the parent because apply_reconstructed_incoming_rows does not exist.

Test passes after fix

$ fo build && fo test helical_source_test
helical_source_test                         PASS       0.11s
Summary: 1 passed, 0 failed, 0 skipped

$ FO_CMAKE_ARGS='-DCMAKE_Fortran_FLAGS=' fo
Static: OK (341 modules, 341 changed, 341 affected)
Build: OK
Tests: OK
Lint: OK
All stages passed (1.8s)

The single-rank HELCORE process fixture was then run through prop_reconstruct=0, 1, and 2 with prop_write=2:

stage 0: 398 propagator files, 198 boundary files, final joined propagator present
stage 1: 199 reconstructed-boundary files, exit 0
stage 2: 199 files in each of phi_mesh, dentf_p/m, spitf_p/m, enetf_p/m,
         and sizeplot_etalev; exit 0
finite scan: 0 NaN/Inf files
trap build: -fcheck=all -ffpe-trap=invalid,zero,overflow; exit 0
runtime marker scan: 0 ERROR/IEEE/SIGFPE/runtime-error matches