In Visual Studio 17.5, Visual Studio will start enforcing signature validation for DAC and DBI. This adds documentation on how to disable this for folks using daily or private builds.

97a51cc regressed the inclusion of source generators in the targeting pack as the `netstandard2.0` inner build was never chosen by the sfx-gen.proj traversal project. Noticed in the SDK consumption PR: dotnet/sdk#29406

This uses the exact same strategy as the unmanaged reader (`Dictionary` with a `lock` around `this`).

RyuJIT started asking about names a lot more than it used to. Pass it UTF-8 pointers from metadata instead of UTF-16-converted, then UTF-8 converted, and pinned byte buffers.

* do not overallocate SocketAddress.Buffer
* make sure CopyAddressSizeIntoBuffer and GetAddressSizeOffset are only accessible as System.Net.Sockets internals on Windows

* Changing the Arm hardware intrinsic tests to be dynamically generated

* Changing the General hardware intrinsic tests to be dynamically generated

* Changing the X86 hardware intrinsic tests to be dynamically generated

* Minor additional cleanup of the hwintrinsic generation files

* Ignore a warning about SelfContained mismatch on an unreferenced project

* Ensure the GenerateHWIntrinsicsTests projects are restored for Mono builds

* UseAppHost=true for GenerateHWIntrinsicTests

* Workaround the issue with parallel builds

* Fixing the VectorDotTest template to correctly compute the sum

* [mono] Add export_symbols_outfile parameter
[mono] Export EntryPoint of methods decorated with UnmanagedCallersOnlyAttribute to specified export_symbols_outfile

* [mono] Create export file and export symbols iff there are specified EntryPoints

* [task] Bootstrap export-symbols-outfile mono aot compiler option to MonoAOTCompiler Task

* Format symbols based on platform

* Address feedback

* Check for exportsymbols when EnableUnmanagedCallersOnlyMethodsExport is true

Not replacing everything because the only reason this has better perf than `ConcurrentDictionary` is that the API surface is totally unergonomic. These are the heaviest hitters.

We should not even be using this parser (#79802). It's surprisingly slow to use this.

This saves about 10% of wallclock time when compiling Hello World (yes, I can't believe my measurements either). We parse the descriptors for the purposes of manifest resources pretty late in the single threaded phase.

When changing the test recently, I have missed the conv.i4 that I should
have removed from the preexisting code. The test was passing when
`runincontext` was passed to the coreclr test run, so I have incorrectly
assumed that it will work in regular runs too.