Reuse lists in token cache filtering logic.

[pmaytak]

Fixes #3178

Changes proposed in this request

• Replace IReadOnlyList with List in accessor and filtering methods (RemoveAll is in concrete List implementation).
• In FilterWithLogging, replace list = list.Where(predicate).ToList(); with list.RemoveAll(e => !predicate(e));
  • Add a boolean parameter whether to filter on the current list or create a new one.
• In other methods, replace Where with RemoveAll and reverse filtering predicate (since instead of keeping the items based on a filter, we now remove items based on the opposite of that filter).

Testing
Existing tests.

Performance impact
Tested with 1 tenant and 10K tokens in cache (tenants don't matter in this scenario). About 40% reduction in memory allocated for client, OBO, and silent calls. Perf improvement is minimal, only about .5 ms for silent call.
Before

Method	CacheSize	Mean	Error	StdDev	Median	Min	Max	Gen 0	Allocated
AcquireTokenForClient	(1, 10000)	5,353.3 μs	411.87 μs	1,207.93 μs	5,424.8 μs	2,887.60 μs	7,847.5 μs	-	1,988 KB
AcquireTokenForOBO	(1, 10000)	6,193.0 μs	341.06 μs	978.57 μs	6,045.6 μs	4,182.15 μs	8,638.4 μs	-	2,256 KB
AcquireTokenSilent	(1, 10000)	5,652.5 μs	378.56 μs	1,086.15 μs	5,605.9 μs	3,390.45 μs	8,480.9 μs	-	2,249 KB
GetAccount	(1, 10000)	184.6 μs	31.33 μs	87.85 μs	152.3 μs	92.75 μs	432.4 μs	-	11 KB
GetAccounts	(1, 10000)	175.9 μs	29.71 μs	81.82 μs	151.8 μs	95.20 μs	431.2 μs	-	11 KB
RemoveAccount	(1, 10000)	10,414.1 μs	211.16 μs	615.97 μs	10,262.3 μs	8,266.30 μs	12,159.5 μs	1000.0000	11,075 KB

After

Method	CacheSize	Mean	Error	StdDev	Median	Min	Max	Gen 0	Allocated
AcquireTokenForClient	(1, 10000)	4,716.3 μs	349.99 μs	1,020.94 μs	4,567.2 μs	2,504.50 μs	6,964.0 μs	-	1,219 KB
AcquireTokenForOBO	(1, 10000)	5,997.8 μs	300.82 μs	882.25 μs	5,898.9 μs	4,146.40 μs	8,369.4 μs	-	1,230 KB
AcquireTokenSilent	(1, 10000)	4,847.9 μs	298.60 μs	861.54 μs	4,939.9 μs	2,915.85 μs	7,107.8 μs	-	1,224 KB
GetAccount	(1, 10000)	167.0 μs	29.82 μs	83.61 μs	129.5 μs	81.10 μs	449.9 μs	-	10 KB
GetAccounts	(1, 10000)	180.4 μs	36.04 μs	102.24 μs	138.6 μs	82.80 μs	514.6 μs	-	10 KB
RemoveAccount	(1, 10000)	10,475.1 μs	376.17 μs	1,085.33 μs	10,465.9 μs	7,783.30 μs	13,363.2 μs	1000.0000	10,562 KB

[bgavrilMS]

@pmaytak - did you invert the 2 perf tables in the description? Because the "before" looks better than the "After".

[bgavrilMS]

Please see the comment around filtering by tenantId first and then by clientid.

[pmaytak]

@pmaytak - did you invert the 2 perf tables in the description? Because the "before" looks better than the "After".

Yes, :) Corrected.