Update label-to-block-lookup takes very long

[oflynne]

Committing labels into a source with blocks missing at lower mipmap levels takes forever at the label-to-block-lookup update. Not sure if this is an artifact of the missing blocks or something else. Needs to be investigated.

[cmhulbert]

I believe the issue here was that when downsampling, the RandomAccessibleInterval over the non-existent blocks still iterates over all entries. This was improved for performance in #546, and in the corresponding imglib2-label-multisets update saalfeldlab/imglib2-label-multisets#34

This instead allows us to skip over cells that are non-existent.