Replace waitFor with async/await

[ntkme]

This PR replaces waitFor with async/await.

There's some trade off in terms of performance, see benchmark below. Personally I think this trade off is worth it as in most of the cases the performance is getting better.

Update: Added a commit to fix regression. Performance is now either nearly equal or way better in every case. The code logic isn't really changed from the first attempt, but dart's compiler seems to have trouble optimize the old one correctly due to how type inference works.

Update 2: still regress a lot in another case, see comment below.

• Fixes Concurrent requests block each other and Stack Overflow under high concurrency dart-sass#1959

Baseline 1.57.1 waitFor:

       user     system      total        real
 1c start up cost  0.000184   0.000801   0.000985 (  0.022070)
1t/100c bootstrap  0.021838   0.021640   0.043478 ( 15.298093)
1t/100000c w/  func w/  url  4.891331   1.683478   6.574809 ( 21.760007)
4t/100000c w/  func w/  url  4.863213   1.674371   6.537584 ( 21.683645)
1t/100000c w/  func w/o url  4.709930   1.602782   6.312712 ( 19.915833)
4t/100000c w/  func w/o url  4.722000   1.590509   6.312509 ( 19.854262)
1t/100000c w/o func w/  url  2.131774   0.488197   2.619971 ( 12.001358)
4t/100000c w/o func w/  url  2.124635   0.487802   2.612437 ( 11.982604)
1t/100000c w/o func w/o url  2.073813   0.497656   2.571469 ( 10.536980)
4t/100000c w/o func w/o url  2.082869   0.506721   2.589590 ( 10.585789)
ruby test.rb  104.23s user 39.80s system 100% cpu 2:23.72 total

This PR 1.57.1 async/await:

       user     system      total        real
 1c start up cost  0.000259   0.000981   0.001240 (  0.024843)           # (smaller is better)
1t/100c bootstrap  0.021841   0.021375   0.043216 ( 15.391264)           # 1.006x 
1t/100000c w/  func w/  url  4.081550   1.060135   5.141685 ( 18.881023) # 0.868x
4t/100000c w/  func w/  url  4.082645   1.058004   5.140649 ( 18.870283) # 0.870x
1t/100000c w/  func w/o url  4.037197   1.063087   5.100284 ( 17.276619) # 0.867x
4t/100000c w/  func w/o url  4.040086   1.064021   5.104107 ( 17.267876) # 0.870x
1t/100000c w/o func w/  url  2.149704   0.511046   2.660750 ( 12.137423) # 1.011x
4t/100000c w/o func w/  url  2.144795   0.512078   2.656873 ( 12.132177) # 1.012x
1t/100000c w/o func w/o url  2.098397   0.505022   2.603419 ( 10.608703) # 1.006x
4t/100000c w/o func w/o url  2.088623   0.505967   2.594590 ( 10.606887) # 1.002x
ruby test.rb  99.62s user 27.62s system 95% cpu 2:13.28 total            # 0.927x

Both tests are compiled locally with same dart version:

pkg-compile-native
  dart compile aot-snapshot bin/dart_sass_embedded.dart -Dversion=1.57.1 -Ddart-version=2.18.6 -Dprotocol-version=1.2.0 -Dcompiler-version=1.57.1 -Dimplementation-version=1.57.1 --output build/dart-sass-embedded.native

Benchmark code:

require_relative './lib/sass-embedded'
require 'benchmark'

c = 100000

Benchmark.bm do |x|
  x.report ' 1c start up cost' do
    Sass.info
  end
  x.report '1t/100c bootstrap' do
    100.times do
      Sass.compile('./node_modules/bootstrap/scss/bootstrap.scss').css
    end
  end
  [true, false].each do |f|
    [true, false].each do |u|
      [1, 4].each do |t|
        x.report "#{t}t/#{c}c #{f ? 'w/ ' : 'w/o'} func #{u ? 'w/ ' : 'w/o'} url" do
          input = f ? 'a {b: foo()}' : 'a {b: c}'
          args = ({})
            .merge!(f ? {functions: {
              'foo()': ->(_) { Sass::Value::String.new('bar') }
            }} : {})
            .merge!(u ? {url: 'file:///tmp/test.scss'} : {})

          t.times do
            threads = []
            threads << Thread.new do
              (c / t).times do
                Sass.compile_string(input, **args).css
              end
            end
            threads.each(&:join)
          end
        end
      end
    end
  end
end

[ntkme]

This actually turned out to regress if a function is passed but never used on a large user input due to most part of the compiler is running in event loop because dart simply doesn't optimize synchronously returned Value for FutureOr<Value> and still put it on event loop: dart-lang/language#2033

[ntkme]

Closing for now.