[release/5.0-preview4] Revert processing bundles in framework (dotnet…

…#35679)

This commit reverts:
Revert "Single-File: Process bundles in the framework (dotnet#34274)"
This reverts commit 78b303d.

Revert "Single-File Bundler: Add a FileSize test (dotnet#35149)"
This reverts commit 779588a.

*Customer Scenario*

Publishing apps as a self-contained single-file doesn't work as expected.

* Publish needs to generate hostpolicy and hostfxr separate from the single file bundle
* Cross-platform publishing is incorrect

*Problem*

Since Static-apphost is not yet ready, processing bundle content in hostpolicy means that  hostpolicy and hostfxr DLLs need to be separate from the bundle. This causes self-contained single-file apps to not be a "single file" temporarily. 

The change also requires supporting changes from the SDK, to publish hostfxr and hostpolicy as separate files, and to invoke HostModel library with arguments that facilitate cross-platform publishing.

*Solution*

To solve these, problem, this change reverts:

Revert "Single-File: Process bundles in the framework (dotnet#34274)" commit 78b303d.

and a dependent test-only change:

Revert "Single-File Bundler: Add a FileSize test (dotnet#35149)" commit 779588a.

*Risk*

Medium
The change is contained to only host components: apphost, hostpolicy, and hostfxr.
However, the change is big, and needs testing in runtime and SDK repos.

*Testing*

Manually tested the SDK by inserting apphost, hostfxr, hostpolicy, and hostmodel library from this build into the `dotnet/packs` preview-4 SDK from dotnet/sdk#11518 build.

Verified that:
* Singlefile apps can be published and run OK for  { Windows, Linux, Osx } x {netcoreapp3.0, netcoreapp3.1, netcoreapp5.0} 
* Cross-targeting builds of single-file apps build and run OK (ex: built on Windos, run on Mac).

Fix WriteLargeJsonToStreamWithoutFlushing test (dotnet#34914)

.NET 5.0.0-preview.1

Fix overaggressive CanBeMadeAtomic check for Set + Notone (dotnet#33409)

We're erroneously converting a set loop to be atomic when it's followed by a notone where the notone's character is in the set.  But if we for example have `[ab]*[^a]`, we can't make the loop atomic, because the `[ab]*` can actually give back something (a `b`) that the `[^a]` will match.  The fix is simply to delete the erroneous, overaggressive checks.