Refactor zarr encode

[jeromekelleher]

No description provided.

[jeromekelleher]

This refactors the encode code-path quite a bit, and opens the way for making this distributable on a cluster. Main changes:

• Merged 1D and 2D encode steps into one, and change rate reporting to bytes
• Add --max-memory for encode (approximate memory budget based on the size of the variant-chunk numpy buffer, seems to work quite well)
• Changed the way we try and keep track of in-progress encoding to adding a "wip_" prefix to all arrays that are being written to. Currently this just waits until the end and renames them all, but the idea is to do this as the array encoding jobs complete.

cc @shz9 @benjeffery