Optimize Deque#concat(Indexable)

[HertzDevil]

Like #13280, but for Deque. Depends on #13257.

The main differences are there were no existing overloads more specialized than Enumerable arguments here, and there is already only a single buffer growth policy. Benchmarks:

Source

require "benchmark"

class Deque(T)
  def concat2(other : Indexable)
    # this PR's implementation
  end
end

{% begin %}
  M = {{ env("M").to_i }}
  N = {{ env("N").to_i }}
{% end %}
puts "M=#{M}, N=#{N}"

x = Deque(Int64).new

puts "Array:"
y_array = Array(Int64).new
Benchmark.ips do |b|
  b.report("ctrl") { x = Deque(Int64).new(M, &.to_i64); y_array = Array(Int64).new(N, &.to_i64) }
  b.report("old")  { x = Deque(Int64).new(M, &.to_i64); y_array = Array(Int64).new(N, &.to_i64); x.concat(y_array) }
  b.report("new")  { x = Deque(Int64).new(M, &.to_i64); y_array = Array(Int64).new(N, &.to_i64); x.concat2(y_array) }
end

puts "Slice:"
y_slice = Slice(Int64).empty
Benchmark.ips do |b|
  b.report("ctrl") { x = Deque(Int64).new(M, &.to_i64); y_slice = Slice(Int64).new(N, &.to_i64) }
  b.report("old")  { x = Deque(Int64).new(M, &.to_i64); y_slice = Slice(Int64).new(N, &.to_i64); x.concat(y_slice) }
  b.report("new")  { x = Deque(Int64).new(M, &.to_i64); y_slice = Slice(Int64).new(N, &.to_i64); x.concat2(y_slice) }
end

puts "StaticArray:"
y_static_array = uninitialized Int64[N]
Benchmark.ips do |b|
  b.report("old")  { x = Deque(Int64).new(M, &.to_i64); x.concat(y_static_array) }
  b.report("new")  { x = Deque(Int64).new(M, &.to_i64); x.concat2(y_static_array) }
end

puts "Deque:"
y_deque = Deque(Int64).new
Benchmark.ips do |b|
  b.report("ctrl") { x = Deque(Int64).new(M, &.to_i64); y_deque = Deque(Int64).new(N, &.to_i64) }
  b.report("old")  { x = Deque(Int64).new(M, &.to_i64); y_deque = Deque(Int64).new(N, &.to_i64); x.concat(y_deque) }
  b.report("new")  { x = Deque(Int64).new(M, &.to_i64); y_deque = Deque(Int64).new(N, &.to_i64); x.concat2(y_deque) }
end

M=1000, N=100
Array:
ctrl 820.27k (  1.22µs) (± 0.95%)  8.89kB/op        fastest
 old 448.18k (  2.23µs) (± 0.72%)  24.5kB/op   1.83× slower
 new 489.15k (  2.04µs) (± 0.92%)  24.5kB/op   1.68× slower
Slice:
ctrl 837.32k (  1.19µs) (± 0.85%)  8.86kB/op        fastest
 old 446.36k (  2.24µs) (± 0.97%)  24.5kB/op   1.88× slower
 new 499.05k (  2.00µs) (± 1.02%)  24.5kB/op   1.68× slower
StaticArray:
old 474.77k (  2.11µs) (± 0.69%)  23.5kB/op   1.13× slower
new 534.29k (  1.87µs) (± 0.96%)  23.5kB/op        fastest
Deque:
ctrl 808.45k (  1.24µs) (± 0.73%)  8.89kB/op        fastest
 old 440.56k (  2.27µs) (± 0.95%)  24.5kB/op   1.84× slower
 new 490.72k (  2.04µs) (± 0.70%)  24.5kB/op   1.65× slower

M=1000, N=1000
Array:
ctrl 493.17k (  2.03µs) (± 1.11%)  15.7kB/op        fastest
 old 192.04k (  5.21µs) (± 0.56%)  31.3kB/op   2.57× slower
 new 345.75k (  2.89µs) (± 2.76%)  31.3kB/op   1.43× slower
Slice:
ctrl 495.68k (  2.02µs) (± 1.07%)  15.7kB/op        fastest
 old 194.21k (  5.15µs) (± 0.53%)  31.3kB/op   2.55× slower
 new 334.78k (  2.99µs) (± 0.75%)  31.3kB/op   1.48× slower
StaticArray:
old 232.06k (  4.31µs) (± 0.49%)  23.5kB/op   1.91× slower
new 444.11k (  2.25µs) (± 0.63%)  23.5kB/op        fastest
Deque:
ctrl 455.51k (  2.20µs) (± 0.84%)  15.7kB/op        fastest
 old 188.27k (  5.31µs) (± 0.68%)  31.3kB/op   2.42× slower
 new 321.27k (  3.11µs) (± 0.56%)  31.3kB/op   1.42× slower

M=100, N=1000
Array:
ctrl 939.99k (  1.06µs) (± 0.90%)  8.89kB/op        fastest
 old 219.48k (  4.56µs) (± 0.72%)  32.8kB/op   4.28× slower
 new 452.07k (  2.21µs) (± 0.70%)  21.4kB/op   2.08× slower
Slice:
ctrl 956.11k (  1.05µs) (± 0.88%)  8.86kB/op        fastest
 old 219.85k (  4.55µs) (± 0.56%)  32.8kB/op   4.35× slower
 new 459.57k (  2.18µs) (± 0.61%)  21.4kB/op   2.08× slower
StaticArray:
old 273.96k (  3.65µs) (± 0.50%)  25.0kB/op   2.51× slower
new 686.58k (  1.46µs) (± 0.66%)  13.6kB/op        fastest
Deque:
ctrl 820.61k (  1.22µs) (± 0.79%)  8.89kB/op        fastest
 old 212.27k (  4.71µs) (± 0.99%)  32.8kB/op   3.87× slower
 new 426.32k (  2.35µs) (± 1.14%)  21.4kB/op   1.92× slower