GPU aggregation layer integration

[macrigiuseppe]

Problem
Aggregation in grid and hexagon layer is currently executed on the CPU with javascript. It runs every time points position changed or size of the bin changes. This causes lag with large datasets

Solution
We will work with the deck.gl team to implement GPU accelerated grid layer. RFC: GPU Aggregation enhancements. GPUGridAggregator will use texture mapping to record aggregation results using multiple weights, and custom blend mode for each aggregation operation. It can use Multiple Render Target feature to perform single draw for aggregating multiple weights

Limitation
GPU Aggregation can only perform MIN, MAX, SUM and MEAN calculation. For other aggregation modes - Median and Mode we will fall back to CPU aggregation.

We also need to fall back to CPU aggregation for color domain calculation, since it is not possible to access GPU aggregation result from javascript.

We propose to use web worker to execute CPU aggregation in the background, avoid UI lagging, and notify the UI of its progress.

[chrisirhc]

Synced with @heshan0131 on this today. We're waiting for a PR on deck.gl to be issued then we'll estimate the work needed on kepler.gl to implement it.

Unfortunately, aggregation capabilities in deck.gl doesn't cover all cases here, so we'll still need to fallback to CPU most of the time.

[nurrochmanmuhammad]

how to perform math between two hexagon aggregation at the same position?