Self-serve chunk loading

[spoutn1k]

Loading individual chunks by loading and unloading them for each use allows for a significant drop in memory usage. For instance, this is the memory profile from the main branch:

With an initial buffer allocation for the image, a subsequent ramp-up to load the entire terrain, and a final buffer allocation for the sub-canvas.

The new behavior skips the loading phase, and the initial buffer allocation:

Both those profiles were calculated on the same world of size ~ 6000x6000.

To accomodate such changes, the entire merging code has been revamped. From there, tiling and file caching will be straightforward to implement.

[spoutn1k]

On the same graph, the gap is more impressive.

That's a 66% memory usage decrease.