Change T1 model selection to be an explicit flag rather than automatic

[alexdaniel654]

Pre-processing steps (registration, EPI distortion correction etc) can often cause quite a lot of slightly negative numbers, this then means the T1 mapping code assumes the data has been magnitude corrected. After multiple people have had issues when trying to debug this behaviour we've decided it makes sense to add a flag when instantiating the class to select if the data should be fit to a magnitude corrected model or absolute model.
I'd suggest the automatic selection is kept as an option (not default) but the threshold for deciding the data has been magnitude corrected (currently if 1% of voxels in the first TI are negative) should be increased so it requires more voxels and ideals a larger magnitude of negative numbers.