Improve sass --embedded performance

[ntkme]

This PR improves the performance by doing the following changes.

• The wire packets of Record type (int, Uint8List) are decoded as streaming.
• Each isolate channel and its dispatcher is persisted rather than recreated for each compilation.
• The communication between main isolate uses Record type (bool, Uint8List), therefore allow a boolean signal to be passed to deactivate the channel at the correct timing, to fix the race condition.
• Use Mailbox via ffi to provide synchronized communication between isolates.

Performance summary:

• For lots of tiny files
  • Single-threaded performance has been improved a lot and getting closer to old single isolate performance.
  • Multi-threaded performance beats all previous versions and even beats multi-processes.
• For high number of round trips (function calls, importers, etc)
  • It beats all previous versions in all running modes (single-threaded, multi-threaded or multi-processes).
• For large files like bootstrap
  • Almost no performance difference in all running modes as this test is CPU & I/O bounded. Multi-processes mode is fastest for any version because multi-threaded mode is bottlenecked on I/O.

[nex3]

I would like to avoid using FFI for this until there's a Dart-Team-owned package available for it. I don't feel confident in my ability to own and maintain that code.

[ntkme]

@mraleph I took your Mailbox implementation and made it work with the current dart-sass. However, we would like to see if Dart Team can own the FFI part and ship it as a package that we can consume. Would you or someone from Dart SDK team be able to help us? Thanks in advance!

Reference: dart-lang/sdk#52121

[mraleph]

@ntkme yes, we are going to provide the package as agreed. I was planning to take a look at this in the next month.

[ntkme]

@nex3 I will split this up into smaller PRs.

[mraleph]

@ntkme we have published https://pub.dev/packages/native_synchronization

[ntkme]

@nex3 Welcome back. This PR is now ready for review.

[nex3]

Thanks for the ping. It may be a few days before I can get around to this, but it's on my radar.

[ntkme]

@nex3 Huge thanks for refactoring the reuse logic and remove some left-overs. I've reverted the commit of bubbling up ProtocolError back to original implementation and updated documentation accordingly.

[nex3]

Thanks!