Remove mjData.qLDiagSqrtInv, add corresponding argument to mj_solveM2.

- `qLDiagSqrtInv` is only required for the dual solvers. It is now computed as-needed rather than unconditionally.
- `mj_solveM2` now requires a new input array `sqrtInvD` which contains the square root of the inverse diagonal D (formerly saved in `qLDiagSqrtInv`).

PiperOrigin-RevId: 710805133
Change-Id: I0622d6a8da3882916824e9c10bad9223c122c321

src/engine/engine_forward.c

@@ -803,7 +803,7 @@ void mj_EulerSkip(const mjModel* m, mjData* d, int skipfactor) {
       }
 
       // factor
-      mj_factorI(m, d, MhB, d->qH, d->qHDiagInv, 0);
+      mj_factorI(m, d, MhB, d->qH, d->qHDiagInv);
     }
 
     // solve
@@ -986,7 +986,7 @@ void mj_implicitSkip(const mjModel* m, mjData* d, int skipfactor) {
       mju_addScl(MhB, d->qM, MhB, -m->opt.timestep, nM);
 
       // factorize
-      mj_factorI(m, d, MhB, d->qH, d->qHDiagInv, NULL);
+      mj_factorI(m, d, MhB, d->qH, d->qHDiagInv);
     }
 
     // solve for qacc: (qM - dt*qDeriv) * qacc = qfrc