Trio-parallel

Features

- Add type hints for `run_sync` (322)
- Use ``tblib`` to enable pickling of tracebacks between processes. Mainly, this preserves context of exceptions including chained exceptions. (323)


Bugfixes

- Prevent Ctrl+C from inducing various leaks and inconsistent states. (239)
- Cleaned up names/qualnames of objects in the trio_parallel namespace. (291)


Deprecations and Removals

- Removed python 3.6 support (236)

Bugfixes

- Fixed a hang on failed worker subprocess spawns that mostly occurred upon accidental multiprocessing recursive spawn. 167
- Fixed a hang on Windows when trying to use `WorkerContext.run_sync` in sequential and concurrent Trio runs. 171


Improved Documentation

- Revamped documentation with tested examples. 168

Updates trove classifiers and loosens CFFI requirement.

With this release I consider the project "feature complete" although, as a beta release still, I would be open to PRs for new features with a strong motivation.

- Added an API to view statistics about a `WorkerContext`, specifically counting `idle_workers` and `running_workers`. 155

- Opportunistically use `cloudpickle` to serialize jobs and results. (115)
- Timeout arguments of `open_worker_context`, `idle_timeout` and `grace_period`,
now work like trio timeouts, accepting any non-negative `float` value. (116)
- Worker process startup is now faster, by importing trio lazily (117)
- `open_worker_context` now returns a context object that can be used to run
functions explicitly in a certain context (`WorkerContext.run_sync`) rather
than implicitly altering the behavior of `trio_parallel.run_sync`. (127)

- Added configuration options for the grace periods permitted to worker caches upon shutdown. This includes a new keyword argument for `cache_scope` and the new top level function `default_shutdown_grace_period`. 108
- `cache_scope` gained a new argument, `init`, and `retire` is no longer called before the first job in the worker. 110