over and over.

Although we cached hashes of `MeshVertexBufferLayout`, we were paying
the cost of `PartialEq` on `InnerMeshVertexBufferLayout` for every
entity, every frame. This patch changes that logic to place
`MeshVertexBufferLayout`s in `Arc`s so that they can be compared and
hashed by pointer. This results in a 28% speedup in the
`queue_material_meshes` phase of `many_cubes`, with frustum culling
disabled.

Additionally, this patch contains two minor changes:

1. This commit flattens the specialized mesh pipeline cache to one level
   of hash tables instead of two. This saves a hash lookup.

2. The example `many_cubes` has been given a `--no-frustum-culling`
   flag, to aid in benchmarking.

table.

The trick with `Borrow` was too clever and ended up actually hashing the
pointer, causing lookups to sometimes spuriously fail. This manifested
itself as a crash in `alien_cake_addict`. Removing the
`MeshVertexBufferLayoutRef` wrapper from the entries in the store
prevents its implementation of `Hash` from being called, solving the
problem.

…layouts