[benchmark] Flatten - Extended test family

The extended `Flatten` test family is based on recently added benchmarks `FlattenLostLoop` and `FlattenListFlatMap`. They had unnecessarily large base workloads, which prevented more precise measurement. Their base workload was lowered by a factor of 20. See discussion on swiftlang#20116 (comment)

Since these are recent additions to the Swift Benchmark Suite, I’m removing the originals and reintroducing them under new names `Flatten.Array.Tuple4.flatMap` and `Flatten.Array.Tuple4.for-in.Reserve`without going through the `legacyFactor`.

Based on these two templates, this commit introduces thorough performance test coverage of the related space including:

* method chain `map.joined`
* naive for-in implementation without `reserveCapacity`
* few Unsafe variants that should serve as aspirational targets for ideally optimized code
* lazy variants
* variants for different underlying types: 4 element Array, struct and class in addition to the original 4 element tuple
* variants that flatten Sequence instead of Array

The tests follow naming convention proposed in swiftlang#20334

Author: palimondo

benchmark/CMakeLists.txt

@@ -84,13 +84,13 @@ set(SWIFT_BENCH_MODULES
     single-source/Exclusivity
     single-source/ExistentialPerformance
     single-source/Fibonacci
-    single-source/FlattenList
+    single-source/Flatten
     single-source/FloatingPointPrinting
     single-source/Hanoi
     single-source/Hash
     single-source/HashQuadratic
     single-source/Histogram
-    single-source/InsertCharacter  
+    single-source/InsertCharacter
     single-source/Integrate
     single-source/IterateData
     single-source/Join