Delete manifest resource cache for NativeAOT

[huoyaoyuan]

Extracted from #109114.

In coreclr implementation, PEAssembly::GetResource also does a linear search for the resource. A typical caller would also cache the returned resource.

The resultant complexity would be O(assembly+resource) not O(assembly*resource), so there's not much concern about time regression.

[dotnet-policy-service]

Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas
See info in area-owners.md if you want to be subscribed.

[jkotas]

there's not much concern about time regression.

Could you please collect some perf numbers so that we understand how much we are giving up here?

[huoyaoyuan]

Could you please collect some perf numbers so that we understand how much we are giving up here?

Testing a single assembly with 1000*10KB resources:

        [Benchmark]
        public string[] GetResourceNames() => typeof(Program).Assembly.GetManifestResourceNames();

        [Benchmark]
        public Stream GetResource567() => typeof(Program).Assembly.GetManifestResourceStream("BenchmarkGround.res.567.bin");

Main:

Method	Mean	Error	StdDev
GetResourceNames	1.835 us	0.0139 us	0.0123 us
GetResource567	2.408 us	0.0468 us	0.0539 us

PR:

Method	Mean	Error	StdDev
GetResourceNames	1.754 us	0.0351 us	0.0311 us
GetResource567	2.342 us	0.0293 us	0.0274 us

It even looks an improvement?

[jkotas]

Testing a single assembly with 1000*10KB resources:

Assembly with 1000 resources is not very typical case. A typical case is <10 resources.

[huoyaoyuan]

For assembly with 20 resource:

main:

Method	Mean	Error	StdDev
GetResourceNames	533.7 ns	2.59 ns	2.42 ns
GetResource10	491.3 ns	3.14 ns	2.79 ns
GetResource20	535.0 ns	2.15 ns	2.01 ns

PR:

Method	Mean	Error	StdDev
GetResourceNames	541.3 ns	4.02 ns	3.76 ns
GetResource10	496.3 ns	3.66 ns	3.43 ns
GetResource20	544.6 ns	3.48 ns	3.08 ns

It looks a very little regression, but not much for typical usages.

[jkotas]

/azp run runtime-nativeaot-outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jkotas]

LGTM. Thank you!