Seeded watersheds implementation for grayscale and affinity images

[hanslovsky]

I have an implementation of the seeded watersheds transform at
https://github.com/hanslovsky/imglib2-algorithm/tree/watersheds-cleanup
comparison with master
The code is pretty clean and documentation will be added before a potential pull request. Right now, I have two concerns:

• Added dependency for primitive type queues:

<groupId>it.unimi.dsi</groupId>
<artifactId>fastutil</artifactId>
<version>7.0.12</version>
</dependency>

• Currently, the output parameter that stores the labeling is a RandomAccessibleInterval. This could be relaxed to be a RandomAccessible. However, that would require a dynamically growing RandomAccessible similar to [GrowingStoreRandomAccessibleSingletonAccess in BigCAT]
  (https://github.com/saalfeldlab/bigcat/blob/ead30af10ea3cd8e56ca784401e7630d63773753/src/main/java/bdv/img/GrowingStoreRandomAccessibleSingletonAccess.java) or a CellRandomAccessible in the style of a LazyCellImg that only populates cells on request.

Please comment on the added dependency issue and feel free to discuss the need for watersheds on unbounded RandomAccessible.

[ctrueden]

/cc @gselzer

[hanslovsky]

FWIW I have an undocumented seeded watershed implementation in saalfeldlab/label-utilities that is also available on the imagej maven repo.

This also means that I will probably not work on adding the seeded watersheds into imglib2-algorithm.