Grid Tiling

[gselzer]

This PR enables a MDA containing a grid plan to present a tiled view within the viewer.

GridLayoutExample.mp4

It is a draft because it:

1. Currently is only implemented for row/column grid plans.
2. It subclasses TensorStoreHandler from pymmcore-plus, and overrides a private method TensorStoreHandler._event_index_to_store_index, which could be unstable long-term.
3. Could be better suited as a contribution upstream (to where, I do not know), and could even share functionality with feat: 2 and 3d async chunked loading pyapp-kit/ndv#22 - opinions @marktsuchida @tlambert03?

[tlambert03]

this is a great feature, and something I agree needs to be implemented somewhere.

Am I correct in understanding that this requires grid members (individual images in a grid) to be shifted exactly a pixel size multiple? and creates a single "virtual" dataset containing all the images?
A an alternative, more general solution, would be to create a new texture for each item in the grid, and let opengl translate it on the canvas as needed based on the exact stage coordinates. (this would allow for subpixel relationships)

[tlambert03]

i made a point of resisting passing a sequence to the handler constructor, with the idea being that handler should be able to deal with multiple sequences (we don't necessarily imply a 1-to-1 mapping of sequence-handler). You can access the currently running sequence via the public TensorStoreHandler.current_sequence (however, I'm realizing now that it's poorly implemented since it will be an attribute error prior to sequenceStarted being called).

[gselzer]

Am I correct in understanding that this requires grid members (individual images in a grid) to be shifted exactly a pixel size multiple? and creates a single "virtual" dataset containing all the images? A an alternative, more general solution, would be to create a new texture for each item in the grid, and let opengl translate it on the canvas as needed based on the exact stage coordinates. (this would allow for subpixel relationships)

Ideally, yes, however we'd need to make some API changes to NDViewer to enable the display of an arbitrary number of datasets, no?

[tlambert03]

correct, that would be a change to ndv