[OrderedDictionary] Deprecate `subscript(offset:)` for now (apple#92)

* [OrderedDictionary] Deprecate index(forKey:), subscript(offset:) and friends for now

* [OrderedDictionary] Update tests

* Update benchmarks

* Un-deprecate index(forKey:)

Add arm64 support to CMake build (apple#64)

This is pretty speculative, I just copied the `SwiftSupport.cmake` from swift-argument-parser which seems to have support for arm64.

[test] checkCollection: Don’t pass decreasing indices to distance(fro…

…m:to:) (apple#63)

Also, try running the conformance checkers on the minimal collection types to catch similar issues.

[Deque][NFC] Remove unused code (apple#49)

Fix nondeterministic compiler crash in debug builds (apple#18)

So as it turns out, the Swift compiler’s MergeModules phase (which SwiftPM uses in Debug builds) has a known issue with dependency cycles between separate source files — it rarely triggers, but when it does, swiftc hits an assertion complaining about a deserialization failure.

The v0.0.1 of Swift Collections happens to include such a cycle, but it only triggers the assertion on a small fraction of machines. I don’t yet know what exactly makes it nondeterministic, but it is reproducible by randomly reshuffling the source filenames of the OrderedSet implementation — a couple dozen randomizations will eventually lead to a state where it happens in every build.

The workaround is to put OrderedSet’s Sequence and RandomAccessCollection conformances in the same file. (OrderedSet uses the standard IndexingIterator, which requires a Collection conformance, while Collection itself refines Sequence — I believe this is the dependency cycle that is causing all this problem.)

Unfortunately, until we figure out what triggers the nondeterminism (or until the MergeModules action is replaced with something that actually, you know, works, this issue may keep randomly reoccurring. None of this makes much sense, but here we are. 😝

Add the benchmarking config used in the announcement post on swift.org