Compile single function in InferenceDataConverter._extract_log_likelihood

[ricardoV94]

After #5245 we no longer need to create one function per observed variable, which may be more efficient.

https://github.com/pymc-devs/pymc/blob/10523ef548d68848d6408f78d52fefea9be5c521/pymc/backends/arviz.py#L266-L271

This requires some understanding of how to feed the right information to the InferenceData object, possibly with some extra room for improvements in InferenceDataConverter.log_likelihood_vals_point which has a TODO as well:

https://github.com/pymc-devs/pymc/blob/10523ef548d68848d6408f78d52fefea9be5c521/pymc/backends/arviz.py#L240-L245