Perf improvements to AcquireTokenForClient flow

[pmaytak]

Issue #2204

Changes include:

• Update GetAllAccessTokens method to return IEnumerable instead of a List.
• Use IEnumerable.Where instead of FilterWithLogging, in ITokenCacheInternal.FindAccessTokenAsync.
• Add checks if the serialization cache delegates were set, before calling them.

Added a BenchmarkDotNet performance test project to test the AcquireTokenForClient. Tested with cache size of 100, 1k, 10k, and 100k items. Also ran the test multiple times. There was some spread in mean data points between the runs (although I'm not sure why). (The spread was wider pre-fixes and much narrower post-fixes, except for .NET 4.8 which basically had no spread.) .NET 5 seems to be slightly faster than .NET Core 3.1 which is slightly faster than .NET 4.8. Below is the summary data.

.NET Core 3.1.9 (CoreCLR 4.700.20.47201, CoreFX 4.700.20.47203)

Items in cache	Mean (pre-fixes)	Mean (post-fixes)	Improvement
100	71 - 91 μs	64 - 73 μs	10 - 20 %
1000	674 - 854 μs	400 - 571 μs	34 - 41 %
10000	7.5 - 10 ms	4.6 - 6.2 ms	38 %
100000	122 - 176 ms	83 - 117 ms	32 - 34 %

Using in-memory token serialization cache yielded very similar results post-fix.
With in-memory token serialization cache:

Items in cache	Mean (post-fixes)
100	66 - 71 μs
1000	407 - 530 μs
10000	5.2 - 6 ms
100000	85 - 114 ms

.NET Core 5.0.0 (CoreCLR 5.0.20.51904, CoreFX 5.0.20.51904)

Items in cache	Mean (pre-fixes)	Mean (post-fixes)	Improvement
100	67 - 89 μs	54 - 62 μs	20 - 31 %
1000	421 - 684 μs	292 - 482 μs	30 %
10000	5.5 - 8.9 ms	4.8 - 5.4 ms	13 - 40 %
100000	111 - 165 ms	69 - 98 ms	38 - 41 %

.NET Framework 4.8 (4.8.4250.0 runtime)

Items in cache	Mean (pre-fixes)	Mean (post-fixes)	Improvement
100	111 - 113 μs	84 - 88 μs	23 - 25 %
1000	593 - 614 μs	515 - 531 μs	14 %
10000	12.3 ms	7.3 ms	41 %
100000	134 - 137 ms	105 ms	22 - 24 %

.NET Framework 4.6.1 (4.8.4250.0 runtime)

Items in cache	Mean (pre-fixes)	Mean (post-fixes)	Improvement
100	105 - 110 μs	78 - 81 μs	26 %
1000	591 - 602 μs	476 - 499 μs	18 - 20 %
10000	12.3 - 12.4 ms	7.4 - 7.9 ms	37 - 40 %
100000	135 ms	100 ms	26 %

Machine specs:

BenchmarkDotNet=v0.12.1, OS=Windows 10.0.19042
Intel Core i7-9850H CPU 2.60GHz, 1 CPU, 12 logical and 6 physical cores
.NET Core SDK=5.0.100

.NET Framework 4.6.1 (4.8.3928.0 runtime)

Items in cache	Mean (pre-fixes)	Mean (post-fixes)	Improvement
100	124 μs	89 μs	29 %
1000	712 - 726 μs	574 - 577 μs	20 %
10000	14.7 ms	8.5 ms	43 %
100000	159 - 162 ms	118 ms	25 - 28 %

.NET Framework 4.6.1 (4.6.1055.0 runtime)

Items in cache	Mean (pre-fixes)	Mean (post-fixes)	Improvement
100	146 - 149 μs	105 μs	29 %
1000	829 - 838 μs	655 μs	21 %
10000	18 - 18.8 ms	10.9 - 11.1 ms	40 %
100000	200 - 205 ms	151 ms	26 %

Machine specs:

BenchmarkDotNet=v0.12.1, OS=Windows 8.1 (6.3.9600.0), VM=Hyper-V
Intel Xeon Platinum 8272CL CPU 2.60GHz, 1 CPU, 8 logical and 4 physical cores
.NET Framework 4.6.1 (4.6.1055.0)

[henrik-me]

Would really like to see numbers for .net 4.6.1 runtime to be able to compare with 4.8 runtime

[henrik-me]

Wondering if we should add a hashmap to make lookup faster?

[jennyf19]

this is much better.

[jennyf19]

+1 to removing code.

[jennyf19]

[pmaytak]

@bgavrilMS As discussed, added ToList in these methods, since we need to investigate what the impact of returning an IEnumerable is.