Stable starlets

[pmelchior]

This PR corrects a problem with the starlet reconstruction algorithm (the treatment of edges was incorrect), which caused edge artifacts. It also removes the proximal operators because they lead to other artifacts across the image in the backward path.

As is, the starlet morphology model is unconstrained and overcomplete, which means you might as well use a 2D image as morphology. However, the solution is to put explicit constraints or priors on the coefficients to encourage some form of sparsity. A notebook describing this option is forthcoming.

[github-actions]

Before [e320124]	After [46e9854]	Ratio	Benchmark (Parameter)
0	0	n/a	bbox_benchmarks.BoxSuite.mem_bbox_contains(16)
0	0	n/a	bbox_benchmarks.BoxSuite.mem_bbox_contains(2)
0	0	n/a	bbox_benchmarks.BoxSuite.mem_bbox_contains(2048)
0	0	n/a	bbox_benchmarks.BoxSuite.mem_bbox_contains(256)
0	0	n/a	bbox_benchmarks.mem_bbox_creation
35.7±0.5μs	35.5±0.4μs	1.00	bbox_benchmarks.BoxSuite.time_bbox_contains(16)
35.6±0.4μs	35.6±0.5μs	1.00	bbox_benchmarks.BoxSuite.time_bbox_contains(2048)
35.6±0.4μs	35.6±0.3μs	1.00	bbox_benchmarks.BoxSuite.time_bbox_contains(256)
7.74±0.01μs	7.77±0.04μs	1.00	bbox_benchmarks.time_bbox_creation
34.5±0.3μs	33.2±0.3μs	0.96	bbox_benchmarks.BoxSuite.time_bbox_contains(2)

Click here to view all benchmarks.

[pmelchior]

@charlotteaward tagging you here so that you know. Anything with a StarletMorphology will now behave like an ordinary unconstrained image, so constraints and/or priors (#247) need to be added to get the expected morphology with larger-scale modes.