perf: Pre-compute harvester raycasts at frame start

[Phantomical]

Before this commit about 1/4 of the runtime of the various harvesters is spent running raycasts. This can also end up being much worse because occasionally a raycast will trigger a transform sync.

Instead of doing this, we can use RaycastCommand.ScheduleBatch to queue all of these up at the start of FixedUpdate, and then individual consumers can look up the result later on when they need it. Consumers subscribe for a single ray trace per part module, with a transform that defines where the array originates from.

The RaycastManager seems to take ~22us with 16 active drills, most of which is spent scheduling the 3 resulting jobs. The actual jobs themselves take about 50us, so this work should scale pretty well to much larger vessels if needed.

Here's what the raycast dispatch looks like. If we ever decide to use burst for other stuff then this could be burst-compiled but otherwise it really doesn't matter.

And here's what an individual converter looks like with this (and KSPModdingLibs/KSPCommunityFixes#374)

Overall test profile for an asteroid harvester

I think this is the final bit needed for #167