Add benchmarks to compare with Base.circshift for simple operations

[piever]

No description provided.

[inkydragon]

Some real world case:

• Julia needs more and more memory as the loop goes on | juliaCN

In the Chinese community of Julia, someone opened a topic complaining that Julia's code took up too much memory.
After the performance test (julia --track-allocation=user filename.jl), I found that Base.circshift took up the most memory.

After some searching, I found this package, and by using this package, the memory consumption problem was solved.

• using Base.circshift: "6.694 s (3507 allocations: 2.55 GiB)"
  allocation count (estimate using --track-allocation flag):
  2617296789 @views f[:, :, k] .= circshift(f[:, :, k], [cx[k], cy[k]])
• using ShiftedArrays.circshift: "5.533 s (3003 allocations: 137.50 MiB)"
  allocation count:
  72000 @views f[:, :, k] .= ShiftedArrays.circshift(f[:, :, k], (cx[k], cy[k]))

reduce 94% memory allocations (estimate using BenchmarkTools.@benchmark)

Some Notes:

• Optimized source code see: 2763v3.0.jl .
  (just copy and run, these codes only needs ShiftedArrays.jl & BenchmarkTools.jl)
• These codes implement the Lattice Boltzmann Method in CFD simulation.