Investigate getting rid of reflection blocking

[MichalStrehovsky]

NativeAOT has a significant amount of the runtime written in C#. Keeping reflection metadata for it results in a lot of garbage being generated by the compiler since nobody is ever going to reflect on these things (and if they do, preventing that sounds like a good thing). We have a concept of reflection blocking for that that we inherited from .NET Native.

But now that IlcTrimMetadata=true is the default, there's little size incentive in keeping this - we should measure the impact. If it's not significant, we should stop doing reflection blocking.

We currently disable reflection blocking in the libs test because they do reflect on CoreLib's implementation details to test them. But it also means we're not running the tests in the configuration that the users will run. It could mask issues such as private reflection being done by the libraries themselves (not just by library tests).

[MichalStrehovsky]

Looked at this briefly.

The first thing we need to do is set the <DebuggerSupport>false</DebuggerSupport> feature switch, or this brings over a bunch of debugging type proxies (that are private, and therefore ignored with reflection blocking).

Once that's addressed, I see a 1.5% size regression for a small app. It's not too bad, but we can do better. A lot of it seems to be native layout related. We have things we can do there to make things better. We should do that first before tackling this.

With IlcTrimMetadata=false, the regression is like 15%, so we definitely need that mode to stick. If we backtrack on that for whatever reason, this is not feasible.