feat: support CDict & DDict for faster operation with repeated dictionary use

[sekiyama58]

Currently the dictionary is setup every time when it is passed to compress(dict:) / decompress(dict:), but the dictionary setup is costly process.
When the dictionary is repeatedly used, creating CDict / DDict in advance makes the compression / decompression much efficient.
This adds the Zstd::CDict & Zstd::DDict class to manage pre-loaded dictionary references.

Benchmark

CDict makes it x~13 times faster to compress spec/user_springmt.json 100 times with spec/dictionary.

                               user     system      total        real
compress with String dict  0.012185   0.001020   0.013205 (  0.013205)
compress with CDict        0.000927   0.000040   0.000967 (  0.000973)

require 'bundler/setup'
require 'zstd-ruby'
require 'benchmark'

data = IO.read("spec/user_springmt.json")
dict = IO.read("spec/dictionary")
cdict = Zstd::CDict.new(dict, 6)
ddict = Zstd::DDict.new(dict)

Benchmark.bm 24 do |x|
  x.report("compress with String dict") do
    100.times { Zstd.compress(data, dict: dict) }
  end
  x.report("compress with CDict") do
    100.times { Zstd.compress(data, dict: cdict) }
  end
end