Modularise observation model

[seabbs]

In trying to fix merge issues with #106 I found that it was quite difficult due to the amount of repeated code across functions. To make ongoing development easier I think we should look at making the observation model parts of the code more modular.

This includes:

• Combining the logit hazards
• Making a true obs function with vectorised and non-vectorised and rng options (precisely how to do this without code duplication is a little tricky)
• Share more code across generated quantities and the main likelihood
• Make a function for producing nowcasts from same data format as used in the main likelihood
• Make a function for producing posterior predictions from the same format as used in the main likelihood
• A vectorised version of expected_obs() that can be reused across functions/use cases.

[seabbs]

All done in #138