Add support for long types

[dlindhol]

We've had a number of datasets where we use longs and it would be a shame to prevent them from hapi access. There is the risk of overflow, however. One that we've ignored in other cases. The hapi spec (https://github.com/hapi-server/data-specification/blob/master/hapi-3.3.1/HAPI-data-access-spec-3.3.1.md) says thjis about data types:

Note that there are only a few supported data types: isotime, string, integer, and double. This is intended to keep the client code simple in terms of dealing with the data stream. However, the spec may be expanded to include other types, such as 4-byte floating-point values (which would be called float), or 2-byte integers (which would be called short).

Not even mentioning 64-bit integers (i.e. longs). We do support floats but since doubles are supported, we only risk precision noise.

Maybe we could convert to 32-bit integers, maybe using fill values for those that exceed the max int?

I still need to test the behavior of the 3 output formats.

[dlindhol]

Should also update the binary codec: https://github.com/latis-data/latis3-hapi/blob/master/src/main/scala/latis/util/hapi/DataCodec.scala#L24

[dlindhol]

This only addresses the ability to serve LaTiS datasets that have variables defined as longs. The ConvertHapiTypes operation will convert these to 32-bit integers (with a risk of overflow'd values) before being sent to the BinaryEncoder so the hapi.DataCodec does not need to deal with longs.

This will also allow LaTiS datasets with long type variables to be in the catalog of supported datasets.

[lindholc]

I am somewhat uneasy about letting things silently overflow.

What if we 1) only converted variables that specify fill/missing/whatever values and use those values if the conversion fails or 2) building on 1, if no fill value is specified, use max int or min int or something and set that as the fill value.

[dlindhol]

Overflow is unfortunately common. Even Scala's toInt does it. I would like to do better. I was thinking of an operation, used as a processing instruction, to convert types where we could centrally handle such cases. We would need the ability to push some operation like time selections before it. That's a bigger task.

The current reality is that we have data sources with variables defined as longs. I see it as a shortcoming that HAPI chose not to include support for 64-bit integers.