Seam accuracy and divergence (M6)

Milestone M6 (issue #68) measures, and then lifts, the accuracy cap of the static 2:1 seams. Phase A adds the instruments (AO, AD, AS, AL); the transfer operators and the reflux policy change only behind new keys, whose defaults keep today's behaviour:

Key	Values (default first)	Phase
[amr] seam_restriction	mean, lagrange6, weno6	C1, C3
[amr] seam_ghost_fill	tricubic, injection, restriction-compatible, quintic, weno6, quintic-mp	D1, D2
[numerics] reflux_policy	always, troubled	E1, E2
[diagnostics] divb_flux	.false., .true.	A2

The investigations behind the phases, and the decisions they led to, are the seam studies.

Every case runs on 2 ranks from its check.sh under src/tests/flume/verification/amr-*/; FLUME_EXE selects the backend. No M6 case has a figure yet: they are listed in #85.

Case	ID	What it verifies
Order across static 2:1 seams	AO (A1, C1)	the order of a smooth vortex crossing a refined quadrant, per region, against the uniform control; the 6-point restriction
Divergence across seams	AD (A2)	D_flux, the divergence the cleaning acts on: bounded and first-order convergent at the seams; #78 is not a divergence difference
Shock crossing a static 2:1 face	AS (A3, C3, C3b, C3c)	mass, shock position and the plateau behind a shock crossing a 2:1 face, against the exact solution; the 6-point and WENO restrictions at shocks
Integral drift and the WENO sensor near seams	AL (A4)	the drift of the integrals without the reflux, and what the WENO sensor reads at the seams