While looking into #192 and #193 I noticed the reference counting of SplatAccumulators wasn't always correct. There were several ways it could go wrong, resulting in a range of issues. Luckily the resulting changeset is small, but tracking them down wasn't so easy. With these changes the ref counting should work correctly. There's still some possibilities for viewpoints to keep hold of accumulators starving the SparkRenderer.

The following changes were made:

• Initialize the first accumulator with a refCount of 1, as the SparkRenderer immediately uses it as its active accumulator
• Determine originChanged between current viewToWorld and the viewToWorld of the newly retrieved accumulator. Previously this compared to the active accumulator, but the one retrieved from the pool might be older
• Let SparkViewpoint keep track of its two references (pending.accumulator and display.accumulator). Before it kept a maximum of 1 reference, but since sorting happens asynchronous there could be in-flight sorting operations using an accumulator that was already released.
• When the new accumulator has correspondence with the previous one, also update the viewToWorld and trigger a prepareViewpoint. Otherwise there'd be a window for a SparkViewpoint to have mismatched accumulator and viewToWorld