Small fix of range correction bug

[mgovorcin]

Description:
Range correction (mean phase shift from wrapped phase) sometimes gives opposite correction sign (skips the cycle) that affects the calculation of the shift between overlapping regions, e.g. instead of subtracting 3 radians it adds them up.

This happens in few cases, most of the time works well. I am not sure why this happens in the first place (could it be the ESD applied on one frame and another no?)

Solution:
I found it out that n integer of 2pi jumps with range correction most of the times is very close to the median difference. So if the difference between those two is larger than a PI, then I switch the sign of range correction.

NOTE: still testing if the median difference between connected components is the most robust option for stitching to avoid the above mentioned.

Another changes:
Added check to accept dem and mask input as filename string instead of gdal.object to be able to use dask to run code in parallel.

[sssangha]

Hello @mgovorcin this LGTM. The velocity field derived from the pairs spanning SF bay through 2019 (the subject of the ariaTSsetup notebook) looks good:

I made the sequential stitcher the default. Let me know if there are any further changes you anticipate for now. Otherwise I will merge and thereafter issue a new release.

[sssangha]

LGTM