`nan` grads when running `find_MAP()` on `analytic`, `ddm`

Something as simple as:
```
# Test 1: Basic DDM model
cav_data = hssm.load_data("cavanagh_theta")

mCon=hssm.HSSM(
data=cav_data,
model="ddm",
)

mCon.find_MAP()
```

Produces `nan` grads (as per report from the optimizer). This may be the key to investigating what is going on with the gradients of that model and therefore corresponding sampler pathologies.