[stdlib] add withContiguousStorage to CollectionOfOne

[oxy]

As it says on the tin - conform to withContiguousStorageIfAvailable in CollectionOfOne. Enables (in theory) being able to take fast paths on certain operations.

[oxy]

@swift-ci please test

[oxy]

@swift-ci please test

[oxy]

@swift-ci please smoke test

[oxy]

Failed Tests (13):
Swift(macosx-x86_64) :: Interop/Cxx/stdlib/overlay/std-string-overlay.swift
Swift(macosx-x86_64) :: Prototypes/UnicodeDecoders.swift
Swift(macosx-x86_64) :: api-digester/stability-stdlib-abi-with-asserts.test
Swift(macosx-x86_64) :: stdlib/NSSlowString.swift
Swift(macosx-x86_64) :: stdlib/SmallString.swift
Swift(macosx-x86_64) :: stdlib/StringAPI.swift
Swift(macosx-x86_64) :: stdlib/StringBridge.swift
Swift(macosx-x86_64) :: stdlib/StringCreate.swift
Swift(macosx-x86_64) :: stdlib/StringIndex.swift
Swift-validation(macosx-x86_64) :: stdlib/String.swift
Swift-validation(macosx-x86_64) :: stdlib/StringUTF8.swift
Swift-validation(macosx-x86_64) :: stdlib/Unicode.swift.gyb
Swift-validation(macosx-x86_64) :: stdlib/UnicodeUTFEncoders.swift

[oxy]

The test failures were caused by _UIntBuffer creating another instance of itself in replaceSubrange combined with #65778, which forwarded from append(contentsOf:) back to replaceSubrange when there's contiguous storage available.

More detailed backtrace:

_UIntBuffer.replaceSubrange creates a new buffer from that forwards to the RangeReplaceableCollection.init<S: Sequence> method:

-> 212 	    let replacement1 = _UIntBuffer(replacement)

which then forwards to RangeReplaceableCollection.append(contentsOf:)

   409 	  public init<S: Sequence>(_ elements: S)
   410 	    where S.Element == Element {
   411 	    self.init()
-> 412 	    append(contentsOf: elements)
   413 	  }

which then forwards to withContiguousStorageIfAvailable -> replaceSubrange pre-revert.

   457 	    let done:Void? = newElements.withContiguousStorageIfAvailable {
-> 458 	      replaceSubrange(endIndex..<endIndex, with: $0)
   459 	    }

which led to infinite recursion, causing crashes in the tests above.

Since the change is now reverted in main (see #67842), tests should now pass / merging should be safe. PR is rebased.

[oxy]

@swift-ci please test

[oxy]

@swift-ci please apple silicon benchmark

[oxy]

Performance (arm64): -O

Improvement	OLD	NEW	DELTA	RATIO
ConvertFloatingPoint.MockFloat64ToDouble	15.837	9.872	-37.7%	1.60x (?)
ConvertFloatingPoint.MockFloat64Exactly	5.148	3.672	-28.7%	1.40x (?)
ConvertFloatingPoint.MockFloat64Exactly2	5.438	3.959	-27.2%	1.37x (?)
ConvertFloatingPoint.MockFloat64ToInt64	391.5	333.5	-14.8%	1.17x (?)
ObserverForwarderStruct	204.615	179.231	-12.4%	1.14x (?)
DataAppendDataSmallToMedium	2069.412	1924.706	-7.0%	1.08x (?)

Code size: -O

Performance (arm64): -Osize

Regression	OLD	NEW	DELTA	RATIO
Set.isDisjoint.Seq.Int.Empty	43.652	49.92	+14.4%	0.87x (?)
Set.isDisjoint.Int.Empty	43.679	49.918	+14.3%	0.88x (?)
Set.isDisjoint.Box.Empty	45.217	51.458	+13.8%	0.88x (?)
Set.subtracting.Seq.Empty.Int	104.444	117.0	+12.0%	0.89x (?)
UTF16Decode.initFromCustom.cont	367.0	408.0	+11.2%	0.90x (?)
Set.subtracting.Seq.Int.Empty	111.111	121.0	+8.9%	0.92x (?)
Set.isStrictSubset.Empty.Int	79.538	85.826	+7.9%	0.93x (?)
Improvement	OLD	NEW	DELTA	RATIO
NaiveRRC.append.smallContiguousRepeated	79.222	72.609	-8.3%	1.09x (?)

Code size: -Osize

Performance (arm64): -Onone

Regression	OLD	NEW	DELTA	RATIO
ArrayAppendToGeneric	204.667	348.0	+70.0%	0.59x (?)
ArraySetElement	562.0	687.0	+22.2%	0.82x (?)
Set.filter.Int100.24k	62.575	69.714	+11.4%	0.90x
Set.filter.Int100.20k	52.585	58.476	+11.2%	0.90x
Set.filter.Int100.16k	42.6	47.231	+10.9%	0.90x
Set.filter.Int100.28k	75.036	83.067	+10.7%	0.90x (?)
String.replaceSubrange.ArrChar	25.414	27.903	+9.8%	0.91x
SIMDReduce.Int8x16.Cast	811.0	886.5	+9.3%	0.91x (?)
Improvement	OLD	NEW	DELTA	RATIO
ArrayAppendGenericStructs	842.308	496.429	-41.1%	1.70x (?)