New and better-behaved shutdown methods for `EventLoopGroupConnectionPool`

[gwynne]

• Soft-deprecate (documentation only, no warnings from the compiler) EventLoopGroupConnectionPool.shutdown() for being unnecessarily synchronous (among other things).
• Add EventLoopGroupConnectionPool.shutdownGracefully(_:), modeled loosely after EventLoopGroup.shutdownGracefully(queue:_:). Shutdown takes place asynchronously, and a callback is invoked on a "safe" Dispatch queue to notify of completion and any errors.
• Add EventLoopGroupConnectionPool.syncShutdownGracefully(), modeled rather closely after EventLoopGroup.syncShutdownGracefully(). Shutdown is synchronous from the point of view of calling code. As with any such method, it must not be invoked on any of the EventLoops in the pool's EventLoopGroup, otherwise deadlock will result. Any error that occurs is thrown.
• Clean up how NIO.Lock is used; now both safer and spends less time with the lock held.
• EventLoopGroupConnectionPool.requestConnection(logger:on:) and EventLoopGroupConnectionPool.withConnection(logger:on:_:) now check for an "pool was already shut down" condition and fail early.
• Add EventLoopGroup.future(result:), a convenience method for using the corresponding version of EventLoopPromise.completeWith(_:). Useful when working with both futures and Results.
• All AsyncKIt unit tests now bootstrap logging and respect the LOG_LEVEL environment variable.
• The EventLoopFutureQueue unit tests now run 4 seconds faster.
• A round of CI improvements, including improvements to the code coverage reporting.

Care has been taken not to break any existing API, despite this resulting in the confusing but for-now unavoidable presence of three distinct methods for shutting down pools. Especial care in particular has been taken to ensure that code using the legacy .shutdown() method will correctly and safely interoperate with code using the new .shutdownGracefully(_:) and .syncShutdownGracefully() methods in all cases.

[codecov-io]

Codecov Report

Merging #64 into master will increase coverage by 0.14%. The diff coverage is 93.95%.

@@            Coverage Diff             @@
##           master      #64      +/-   ##
==========================================
+ Coverage   94.43%   94.58%   +0.14%     
==========================================
  Files          23       23              
  Lines        1043     1182     +139     
==========================================
+ Hits          985     1118     +133     
- Misses         58       64       +6     

Flag	Coverage Δ
#unittests	94.58% <93.95%> (?)

Impacted Files	Coverage Δ
Sources/AsyncKit/EventLoop/EventLoop+Future.swift	60.00% <0.00%> (-40.00%)	:arrow_down:
...es/AsyncKit/EventLoopFuture/FutureExtensions.swift	73.33% <0.00%> (-26.67%)	:arrow_down:
.../ConnectionPool/EventLoopGroupConnectionPool.swift	93.00% <93.33%> (-1.12%)	:arrow_down:
Tests/AsyncKitTests/AsyncKitTests.swift	100.00% <100.00%> (ø)
Tests/AsyncKitTests/ConnectionPoolTests.swift	89.53% <100.00%> (+2.49%)	:arrow_up:
...ests/AsyncKitTests/EventLoopFutureQueueTests.swift	98.61% <100.00%> (+0.02%)	:arrow_up:
Tests/AsyncKitTests/EventLoopGroup+Future.swift	100.00% <100.00%> (ø)
Tests/AsyncKitTests/FlattenTests.swift	100.00% <100.00%> (ø)
Tests/AsyncKitTests/Future+CollectionTests.swift	100.00% <100.00%> (ø)
Tests/AsyncKitTests/Future+OptionalTests.swift	100.00% <100.00%> (ø)
... and 4 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 48a719d...2725124. Read the comment docs.

[gwynne]

One question about CI, how come there is a "push" and "pull_request" run for each image?

This seems like its unnecessarily doubling the amount of test cases that need to run on each PR. Or am I missing something?

Search me. It was part of the configuration when I got here, I assumed originally it had something to do with making codecov work and left it alone. I would be delighted to yank the extra event if it's actually superfluous.

[tanner0101]

Hmm. vapor/vapor only has pull_request: https://github.com/vapor/vapor/blob/master/.github/workflows/test.yml#L3

I think we added push as well so that code cov baseline would get updated when merging to master. Maybe we can just have push and not pull_request, too? Seems really extra to have each test run twice.

[gwynne]

I think we added push as well so that code cov baseline would get updated when merging to master. Maybe we can just have push and not pull_request, too? Seems really extra to have each test run twice.

If the goal is to update the baseline for merges, there's at least one better event for that. I'll do a quick poke through the reference and find an appropriate one.

[tanner0101]

These changes are now available in 1.1.0