Fix: HttpClientFactory: resolving a typed client is significantly slower that resolving a named client.

[rynowak]

See commits for details.

BenchmarkDotNet=v0.10.13, OS=Windows 10.0.18362
Intel Core i7-8650U CPU 1.90GHz (Kaby Lake R), 1 CPU, 8 logical cores and 4 physical cores
.NET Core SDK=3.0.100-preview6-012264
  [Host]     : .NET Core 3.0.0-preview8-27919-01 (CoreCLR 4.700.19.36901, CoreFX 4.700.19.36803), 64bit RyuJIT
  Job-PYIOAJ : .NET Core 3.0.0-preview8-27919-01 (CoreCLR 4.700.19.36901, CoreFX 4.700.19.36803), 64bit RyuJIT

Runtime=Core  Server=True  Toolchain=.NET Core 3.0  
RunStrategy=Throughput  

Before:

Method	Mean	Error	StdDev	Median	Op/s	Scaled	ScaledSD	Gen 0	Allocated
'use IHttpClientFactory with named client'	13.87 us	0.2708 us	0.3521 us	13.84 us	72,116.5	1.23	0.08	7.8125	3.18 KB
'use IHttpClientFactory with typed client'	500.04 us	17.8916 us	52.1905 us	515.59 us	1,999.9	44.26	5.34	-	8.63 KB
'new HttpClient'	11.34 us	0.2386 us	0.6922 us	11.28 us	88,183.9	1.00	0.00	7.8125	2.66 KB

After:

Method	Mean	Error	StdDev	Op/s	Scaled	ScaledSD	Gen 0	Allocated
'use IHttpClientFactory with named client'	11.065 us	0.2203 us	0.4549 us	90,376.9	1.88	0.09	7.8125	3.17 KB
'use IHttpClientFactory with typed client'	13.362 us	0.2660 us	0.6059 us	74,839.8	2.27	0.11	7.8125	3.28 KB
'new HttpClient'	5.877 us	0.1116 us	0.1194 us	170,146.8	1.00	0.00	6.8359	2.67 KB

[rynowak]

/cc @damianh

[BrennanConroy]

From my limited knowledge, this looks good.