*_PDC physics-dynamics coupling diagnostics wrong on CSLAM (pgN) grids: missing dp_fvm weighting

### What happened?

Observed by `claude-fable-5-1` in CAM-SIMA FHIST_C4 ne16 runs but the code is identical in CAM.

The column-integrated physics-dynamics coupling loss diagnostics `WV_PDC, WL_PDC, WI_PDC, TT_PDC` are output in the wrong units on any CSLAM (fv_nphys > 0, pgN) grid. They are too small by roughly a factor of dp_fvm (~1e2 to 1e4 Pa).

This is diagnostic-only and is not model state changing (not sure if it is answer-changing or not depending on regression test coverage). Both `ftmp` and `ftmp_fvm` are flagged "Only used for diagnostics!".

GLL / pg0 output is correct.

**Proposed fix**

Weight the fvm increment by dp_fvm at the source, matching the energy-diagnostics usage, in `prim_advance_mod.F90:385`

```fortran
ftmp_fvm(i,j,k,q,ie) = (tmp-v1)*fvm(ie)%dp_fvm(i,j,k) !Only used for diagnostics!
```

### What are the steps to reproduce the bug?

Run FHIST_C4 ne16pg3: WV_PDC values are <= 0 with magnitude <= 2e-26, consistent with mixing-ratio (kg/kg) units rather than kg/m2.

### What CAM tag were you using?

cam6_4_199

### What machine were you running CAM on?

CGD machine (e.g. izumi)

### What compiler were you using?

GNU

### Path to a case directory, if applicable

_No response_

### Will you be addressing this bug yourself?

Any CAM SE can do this

### Extra info

Root cause analysis by `claude-opus-4-8` (reported originally by `claude-fable-5-1`)

`util_function` (`src/dynamics/se/dycore/prim_advance_mod.F90`) column-integrates its input and multiplies by `rga` (1/g), so it expects a `Qdp`-like mass in Pa*kg/kg and returns kg/m2.

- GLL path: `ftmp = dt*FQ - v1` is a Qdp increment (Pa*kg/kg), since `v1 = dt*FQ` is added directly to state%Qdp. Correct.
- CSLAM path: `tmp = dt*fc/dp_fvm`, and `fc = fc_phys*dp_fvm` (fvm_mapping.F90), so `tmp = dt*fc_phys` is a mixing-ratio increment (kg/kg), applied to `fvm%c`. Therefore `ftmp_fvm = tmp - v1` is kg/kg, missing the dp_fvm factor.

`output_qdp_var_dynamics` passes both to util_function identically, so the CSLAM diagnostic is short one dp_fvm.

Corroboration: in the energy/mass diagnostics in the same file, every CSLAM `fvm%c` handed to util_function is first multiplied by dp_fvm (e.g. `cdp_fvm = fvm(ie)%c(...)*fvm(ie)%dp_fvm(...)`), while the GLL branch passes `qdp` directly. That is exactly the factor absent from the PDC path.