Implement TakeLast performance optimizations and benchmarks

[Akeit0]

#204
Simple TakeLast(100) Benchmark.
I wonder if optimization by TryGetSpan is needed.
(In fact, it might have been better if ValueEnumerable had an InnerSkip functionality)

Method	N	Mean	Error	StdDev	Gen0	Allocated
FromArray_Linq	1000000	228.8 ns	33.15 ns	1.82 ns	0.0086	136 B
FromArray_ZLinq	1000000	3,547,090.8 ns	171,427.16 ns	9,396.50 ns	-	48 B
FromArray_ZLinq(PR)	1000000	469,546.8 ns	162,486.33 ns	8,906.42 ns	-	-
FromList_Linq	1000000	242.9 ns	28.69 ns	1.57 ns	0.0086	136 B
FromList_ZLinq	1000000	3,547,946.9 ns	122,781.05 ns	6,730.04 ns	-	48 B
FromList_ZLinq(PR)	1000000	545,908.3 ns	4,946.87 ns	271.15 ns	-	-
FromEnumerable_Linq	1000000	4,456,278.1 ns	76,890.24 ns	4,214.61 ns	-	1368 B
FromEnumerable_ZLinq	1000000	4,021,346.6 ns	164,733.85 ns	9,029.62 ns	-	136 B
FromEnumerable_ZLinq(PR)	1000000	3,410,969.7 ns	1,775,177.79 ns	97,303.49 ns	-	128 B

[Akeit0]

Hmm... How should the test be passed?
BinaryOperations_SourceEnumeratorsDisposed
System.OutOfMemoryException : Array dimensions exceeded supported range.
I don't think this is a good way.

var totalCount = 0X7FFFFFC7;//Array.MaxLength
if (source.TryGetNonEnumeratedCount(out totalCount) ||0X7FFFFFC7 < takeCount)

[neuecc]

This isn't about errors, but I'm generally avoiding having enumerators hold pools directly as fields (for example, OrderBy and Reverse deliberately wrap them https://github.com/Cysharp/ZLinq/blob/9218fc6c/src/ZLinq/Internal/RentedArrayBox.cs#L5-L6)
ValueRingBuffer might be going against this policy.

[Akeit0]

I fixed the error and make RingBuffer class.

The test error happening on ZLinq.Tests.Linq.OrderBySkipTakeTest.OrderByTake is very strange. It has nothing to do with this PR and passes in my local environment.

edit
Since I missed array clear?

[Akeit0]

I think this PR is ready to merge.

[Akeit0]

Should we take over this behavior?

https://github.com/dotnet/runtime/blob/5535e31a712343a63f5d7d796cd874e563e5ac14/src/libraries/System.Linq/src/System/Linq/Take.cs#L99