Optimize patch operations

There are three simple optimizations:
- Replace the HashMap in the configuration for a BTreeMap for a slightly reduced memory footprint.
- For the patches, when can aggregate them into a single operation instead of multiple operations
- This is specially useful if the pod has no nodeSelector or tolerations at all, since we completely avoid a loop
- For the empty object or empty list, the behavior is similar to the one above

This could easily be benchmarked in a test or two, to validate the results