Decompose conservatively-masked data, but denoise liberally-masked data

[tsalo]

Summary

Per today's call, we should try to adapt the denoising workflow so that the decision tree is applied to the conservatively-masked combined data (i.e., an adaptive mask threshold of 3), but a liberally-masked version is what's denoised (i.e., an adaptive mask threshold of 1).

Additional Detail

This should address concerns about coverage brought up in #600 and #680.

Next Steps

1. Estimate T2*/S0 with liberal threshold.
2. Perform optimal combination with both thresholds.
3. Feed conservatively-masked version into the decomposition and decision tree portions of the tedana workflow.
4. Feed liberally-masked version into the actual denoising portion of the workflow.
5. Output both versions of the combined data, as well as T2* and S0 maps.