Improve Zstd.decompress performance by avoiding unnecessary memory copy

[Watson1978]

Currently, rb_decompress allocates a temporary buffer and copies the input string using ALLOC_N and memcpy. This overhead is significant, especially for large data, and doubles the memory usage during decompression.

benchmark

require 'bundler/inline'
gemfile do
  source 'https://rubygems.org'
  gem 'zstd-ruby'
  gem 'benchmark-ips'
end

require 'securerandom'
large_data = SecureRandom.random_bytes(1024 * 1024 * 100)
compressed_data = Zstd.compress(large_data)

Benchmark.ips do |x|
  x.time = 15

  x.report("zstd decompress") do
    Zstd.decompress(compressed_data)
  end
end

before

$ ruby zstd.rb
ruby 3.4.8 (2025-12-17 revision 995b59f666) +PRISM [x86_64-linux]
Warming up --------------------------------------
     zstd decompress     1.000 i/100ms
Calculating -------------------------------------
     zstd decompress     11.148 (± 9.0%) i/s   (89.70 ms/i) -    167.000 in  15.085672s

after

$ ruby zstd.rb
ruby 3.4.8 (2025-12-17 revision 995b59f666) +PRISM [x86_64-linux]
Warming up --------------------------------------
     zstd decompress     2.000 i/100ms
Calculating -------------------------------------
     zstd decompress     20.890 (± 9.6%) i/s   (47.87 ms/i) -    312.000 in  15.015814s

[SpringMT]

Thank you！