Optimize stub_with

[ypconstante]

Today stub_with calls stub for each functions that needs to be stubbed. This causes a significant overhead because some validations being done repeatedly, and sending multiple messages to store the state change.

This PR updates stub_with to avoid the issues above by storing a list with the stub key and expectation, and sending a single state update.

Name                   ips        average  deviation         median         99th %
bench (PR)         33.02 K       30.28 μs    ±25.30%       28.40 μs       65.52 μs
bench (main)        2.88 K      347.63 μs     ±9.18%      339.18 μs      453.12 μs

Comparison: 
bench (PR)         33.02 K
bench (main)        2.88 K - 11.48x slower +317.34 μs

Memory usage statistics:

Name                 average  deviation         median         99th %
bench (PR)           7.53 KB     ±0.15%        7.53 KB        7.53 KB
bench (main)        55.34 KB     ±0.10%       55.36 KB       55.36 KB

Comparison: 
bench (PR)           7.53 KB
bench (main)        55.34 KB - 7.35x memory usage +47.81 KB

defmodule BenchBehaviour do
  @functions Enum.map(1..50, &:"get_v#{&1}")

  for function <- @functions do
    @callback unquote(function)() :: integer()
  end
end

defmodule BenchBehaviourImpl do
  @behaviour BenchBehaviour

  @functions Enum.map(1..50, &:"get_v#{&1}")

  for function <- @functions do
    @impl BenchBehaviour
    def unquote(function)() do
      unquote(function)
    end
  end
end

Mox.defmock(BenchBehaviourMock, for: BenchBehaviour)

Benchee.run(
  %{
    "bench" => fn ->
      Mox.stub_with(BenchBehaviourMock, BenchBehaviourImpl)
      NimbleOwnership.cleanup_owner({:global, Mox.Server}, self())
    end
  },
  time: 10,
  memory_time: 2
)

[josevalim]

💚 💙 💜 💛 ❤️