There seem to be some misunderstandings. Your data is 1d, but you have two sensors, so you are actually in a multivariate case. If you specifically want to do Kalman filtering, you need to write down a linear system that explains how the senor measurements (the states of the system) are combined into a single observation. The code I linked shows how to implement this, and the book you linked explains how to write down a linear state space. That said, Kalman filtering is very restrictive. The state system must be linear, and all the noise must be Gaussian. You appear to want non-Gaussian noise, which means you need to use an extended Kalman filter. This is non-trivial.
I think your problem is actually much simpler. You can estimate a single weighting parameter to join the measurements of the two sensors, and use the resulting fusion as the mean of your distribution:
# mean of 0.75, for a strong prior bias towards the "strong" sensor
rho = pm.Beta('senor_1_weight', alpha=6, beta=2)
mu = rho * sensor_1_data + (1 - rho) * sensor_2_data
sigma = pm.HalfNormal('obs_sigma')
obs = Hypsecant('obs', mu=mu, sigma=sigma, observed=data)
You could further refine this by adding sensor noise, for example by adding likelihood functions for sensor_1_data and sensor_2_data, maybe as guassian random walks. You could then estimate noise parameters for each of those.