Add bad data filtering and non-standard name mappings to preprocessing

[stevehadd]

Initial data exploration has shown some previously unknown issues with the data. There are not many but they do cause problems for the explorations and classification code that are not easily resolved automatically. For now therefore we can simply exclude observations with these issues, so the code runs. Current issues:

• bad dates - some dates have a bad day of the month recorded
• negative depths - some profiles have a negative depth recorded. It is not clear what this means and further clarification has been sought from domain experts. For now we can exclude these, but in future we may do additional processing based on advice from the ocean scientists.
• some of the labels don't match the standard model names, for example "XBT-4". Is this a different probe model, or should this be labelled a T4? We will need to check with the ocean scientists and come up with a mapping from the given labelled to the standard labels.

[stevehadd]

Not sure when precisely this happened, but the dataset object is now dealing with these problems, mostly by just excluding the data.