Box TokioTasksRuntime for efficient moves into and out of world during tick_runtime_update

[JoeHowarth]

Optimizing for this move every frame:
https://github.com/EkardNT/bevy-tokio-tasks/pull/7/files#diff-b1a35a68f14e696205874893c07fd24fdb88882b47c23cc0e0c80a30c7d53759R82-R85

[EkardNT]

The current TokioTasksRuntime struct is 120 bytes on my 64-bit machine, which does seem to be worth the tradeoff of one additional indirection when accessing the proposed new inner field.

But, from an encapsulation point of view, I would prefer not to expose this internal optimization to users of the crate. So I suggest making the TokioTasksRuntimeInner struct not pub, which will also require removing the pub on the TokioTaskRuntime tuple struct entry and the runtime field inside the TokioTasksRuntimeInner. Then replace the ability to directly access the pub runtime field with a public runtime() method on TokioTasksRuntime which returns a reference to the Runtime. This will also require updating the methods already on TokioTasksRuntime to have a bunch of self.0 references, of course.

[EkardNT]

I am going to merge this to get the basic change in and then tweak a few details as described above.

[EkardNT]

This improvement is now available in 0.9.5.