Missing measures

[visr]

For measures, floats smaller than -1e38 are considered "no data".

Perhaps we should convert these to missing, and write missing to small floats.
This also goes for the Mmin and Mmax in the header, which will be missing if all measures are missing. If this is the right interpretation of the spec that is.

Came up in #25

[rafaqz]

Another option is to use a sentinel approach with a wrapper array type, and replace the missing values on the fly in getindex. This may have less performance cost for the cases where you don't actually use the measures data.

[visr]

Yeah that could work as well. I'm not aware of anyone having needed support for this yet, I don't know how common it is. Perhaps we can even get away with just documenting it, and/or reading as NaN.