Benchmark use of `np.where()` versus boolean masking in Parcels kernels and interpolators

[erikvansebille]

When starting to vectorise the particle loop in Parcels-v4, I chose to use/rely on np.where() statements as a replacement for if-statements. However it could be that boolean masking is much more efficient or cleaner. This could be explored in future benchmarking?

Originally posted by @VeckoTheGecko in #2152 (comment)