test: drop dangling references to the out-of-repo DM/DMK/DMR planning doc

The check_extra.py scripts of the T4-T7 integration cases referenced
docs/dm_dmk_dmr_action_plan.md, which is intentionally kept out of the
repository. Replace those pointers with self-contained text; no behavior
change (verified with python3 -m py_compile).

Author: dyzheng

tests/05_rtTDDFT/18_NO_hyb_TDDFT/check_extra.py

@@ -23,8 +23,8 @@
 
 NOTE on sensitivity: the current is a closed Frobenius trace with a full R sum,
 so it is protected against the e^{-ikR}/e^{+ikR} DMR phase choice (flipping the
-sign leaves all keys at roundoff; see docs/dm_dmk_dmr_action_plan.md). These
-checks guard against gross pairing/phase breakage (e.g. a mismatched
+sign leaves all keys at roundoff). These checks guard against gross
+pairing/phase breakage (e.g. a mismatched
 phase_hybrid in the hybrid gauge). The DMR sign sentinel is
 DMTest.T1_fourier_round_trip.
 

tests/integrate/240_NO_KP_15_SO_FD/check_extra.py

@@ -20,9 +20,8 @@
 
 NOTE on sensitivity: this non-magnetic GaAs SOC case exercises the nonlocal
 SOC force path, but its force is (numerically) identical under the
-e^{-ikR}/e^{+ikR} DMR phase choice (closed-trace protection, see
-docs/dm_dmk_dmr_action_plan.md). The sentinel for the DMR Fourier sign itself
-is the unit test DMTest.T1_fourier_round_trip
+e^{-ikR}/e^{+ikR} DMR phase choice (closed-trace protection). The sentinel
+for the DMR Fourier sign itself is the unit test DMTest.T1_fourier_round_trip
 (source/source_estate/module_dm/test/test_cal_dm_R.cpp), which turns red when
 the sign is flipped. This case guards FD-consistency of the analytic force,
 including the SOC nonlocal contribution.

tests/integrate/260_NO_DJ_PK_PU_AFM_URAMPING_MUL/check_extra.py

@@ -8,8 +8,7 @@
 magnetizations of comparable magnitude, the O sites must stay (almost)
 non-magnetic, and the spin-channel charges must add up to the total charge.
 Checking the channels separately (rather than only the total charge/force)
-guards against an accidental cancellation of the two channel errors, as
-warned in docs/dm_dmk_dmr_action_plan.md section 8.
+guards against an accidental cancellation of the two channel errors.
 
 Each key is a pass/fail flag: 0.0 = pass, 1.0 = fail. Measured values are
 written to stderr for diagnostics.

tests/integrate/260_NO_DJ_PK_PU_SO_4K_MUL/check_extra.py

@@ -8,10 +8,10 @@
    (OUT.autotest/running_scf.log, "Total magnetism (Bohr mag/cell)"),
    both in magnitude and direction.
 
-This is the calibration called for in docs/dm_dmk_dmr_action_plan.md T6:
-for nspin=4 and multi-k the Mulliken m_y path uses S^T instead of S^dagger, so
-an anomalous m_y deviation (order ||Im S(k)||) would flag the B-class gemm
-change. Measured deviation on this case (Fe, 4x1x1, U=5 eV, converged):
+This calibration targets the nspin=4 multi-k Mulliken m_y path, which uses
+S^T instead of S^dagger: an anomalous m_y deviation (order ||Im S(k)||) would
+flag the B-class gemm change. Measured deviation on this case (Fe, 4x1x1,
+U=5 eV, converged):
   |sum Mulliken - rho_cell| = (0.042, 0.004, 0.044) uB on |m| ~ 5.5 uB/cell,
   i.e. the small components agree as well as the large ones, so no anomalous
   m_y deviation is present and the B-class change is NOT executed.