Investigate refactoring `model_tides` to improve parallelisation

[robbibt]

Describe your idea or suggestion
Possible ideas:

• Crop and load consistituents first
• Reuse constituents data and only parallelise interpolation
• Return consistuent values as an xarray? e.g. constituents as data variables with x and y dims
• Perform tide modelling outside loop