Remap nans in the RoiDescriptor TP convertor

Some algorithms mistakenly write nans into the bytestream, which cuses FPEs when the RoiDescriptor is constructed from them.

This remaps the values to physicsal values - ie it creates a 0 width Roi is both limits are nan.

It writes a warning - using the old log output since the code was already using that, and I wanted to be consistent - but otherwise does nothing, so doesn;t print out what the collection is called that is trying to be built, and should hopefully not report anything later.

This is rather dangerous since it might mask nan being produced in the limit for some other reason, so this is perhaps not really ideal.

Not sure whether throwing an std::invalid_argumnent is the correct thing to do - maybe ATLAS has a exception type that should be used

Partially addresses ATR-22687