Unify handling of the @ character in razor code blocks

[333fred]

We were extremely inconsistent with how @ was handled in razor code blocks. When @ was not followed immediately by a : or <, we would enter a recovery loop that would eat all characters until the next {, }, or <; this is not correct, and resulted in a number of inconsistent behaviors as well as bad code. In order to make the existing behavior and new behavior difference understandable, I've done a few different test refactorings, so I would highly recommend reviewing this commit-by-commit. Commit summaries:

1. Have everything base whether to generate baselines on file. The source generator tests still use something different (Conditional methods), but we can address that later.
2. Have all tests verify their C# diagnostics. There were several tests that are affected by this @ change where it wasn't clear that they weren't actually intended to compile. Now, all integration tests will verify that their C# diagnostics. We should at some point do the same for Razor diagnostics, but we may need to make a similar verification framework for razor diagnostics like we have for C# diagnostics.
3. Add new tests for @ scenarios to show how it currently behaves.
4. Finally, the real change: follow the expected rules for @, and update the test baselines to show the new behavior.

Here are the expected rules for @:

1. When at the start of a statement, @ indicates a transition into an embedded render expression.
2. When inside a statement:
   1. @: or @< indicate a transition into a render template
   2. Otherwise, we look for a single identifier after @, as C# would

Fixes #10186.

[333fred]

@dotnet/razor-compiler modulo some possible test failures that I haven't seen outside the compiler layer yet, I believe this should be ready for review. Please give it a look.

[chsienki]

Love it. LGTM.

[333fred]

@chsienki @jjonescz I believe I've addressed feedback, please take another look.