Warn about precision loss when extracting values

[dgobbi]

The vtkDICOMValue class has AsInt(), AsDouble(), AsString() etc. methods for extracting the value of a DICOM data element. This extraction can result in loss of precision. For example:

• extracting 'DS' with AsDouble() or AsFloat() is inexact (decimal mantissa to binary mantissa)
• extracting 'DS' with AsInt() results in truncation

Of course, the only lossless way to extract 'DS' is with AsString(), but that is not useful when the value is to be used as a number.

For each VR, each conversion can be characterized as:

• exact
• inexact (DS to floating-point)
• loss of precision (truncation, or insufficient bits)
• signed to unsigned or vice-versa
• nonsensical (value not representable after conversion)

Whenever a conversion is not exact, it might be good to have a warning produced at runtime if the code is compiled in debug mode. The signed/unsigned is tricky because sometimes DICOM is hazy on 'US' vs. 'SS', e.g. for implicit VR it can depend on PixelRepresentation. Also, the first/third values in the lookup table descriptor are always to be interpreted as unsigned even if the VR is 'SS'.

[dgobbi]

Of course, it would be better to report precision loss at compile time. For example, GDCM uses a template to map tags to VRs and from there to type and encoding: 'DS' has type 'double' and encoding 'ascii'. Philosophically, this is very different from vtk-dicom where the dictionary is simply a data structure. Architecturally, vtk-dicom uses metaprogramming only in its implementation, never in its interface.