Add LINELET regression case for laminar NACA0012

[Syphonicc]

Addresses the regression-test part of #2877.

Adds a serial regression case for LINEAR_SOLVER_PREC= LINELET. There is currently no LINELET coverage in TestCases.

TestCases/navierstokes/naca0012/lam_NACA0012_LINELET.cfg is the laminar NACA0012 config with one added line, LINEAR_SOLVER_PREC= LINELET. The baseline uses the default ILU(0).

Note: it is based on the updated lam_NACA0012.cfg from #2866 (ITER instead of the deprecated EXT_ITER, CFL 30), so on current develop the two files differ in more than that one line until #2866 is merged. This PR does not depend on #2866; the new file parses on develop as-is. The registration is placed after cylinder_lowmach, away from the lam_naca0012 entry added in #2866, so the two PRs don't touch the same lines.

At test_iter = 20 the two cases already differ, so the test is sensitive to the preconditioner:

Preconditioner	rms[Rho]	rms[RhoU]	rms[RhoV]	rms[RhoE]
ILU(0) (lam_NACA0012)	-5.522140	-3.127430	-3.286926	-0.109971
LINELET	-5.282867	-2.845208	-2.852421	0.154542

test_vals are from a local serial clean build of current develop (8118ee0, GCC 15.2.1). The same procedure reproduces the #2866 lam_naca0012 values exactly. I'll update them if CI differs.

This does not address the segfault reports in #2877, and only covers the serial path.

[bigfooted]

can you check parallel?

[Syphonicc]

Good catch LINELET segfaulted in parallel with ≥4 ranks (1 and 2 ranks ran fine, which is why it didn't show up before).

Cause

• Partitions without wall markers have no linelets, so CSysMatrix::BuildLineletPreconditioner never allocated the per-thread working vectors but still indexed them → segfault (CSysMatrix.cpp, confirmed with gdb on the rank that had no walls).
• CGeometry::GetLineletInfo used linelets.empty() as its "already computed" check. On ranks with no linelets it recomputed on every call, including the MPI_Allreduce/MPI_Allgather, while the other ranks returned early → mismatched collectives.

Fix

• Always allocate the linelet working vectors (they're small, MAX_LINELET_POINTS per thread).
• Added an explicit computed flag to CLineletInfo so every rank computes it exactly once.
• Skip the "points in each line" message when there are no linelets globally (avoids a divide by zero).

Tests

• Added lam_naca0012_linelet to parallel_regression.py with mpirun -n 4. The default 2 ranks doesn't trigger the bug, because both partitions touch the airfoil.
• Merged develop and updated the serial test_vals.
• Checked the case runs at 1, 2, 4 and 8 ranks.

Fixes #2877.