C# completion list is currently split across three "rewriter" objects that are async. However, none of them need to be async if they're passed a RazorCodeDocument. So, this change acquires a RazorCodeDocument up front before calling the rewriters. With this in place, the rewriters can be synchronous.

Now the completion list rewriters are synchronous, the RazorCodeDocument can be acquired by callers of RewriteCSharpResponseAsync and passed to it. Then, RewriteCSharpResponseAsync can be made synchronous.

The only reason this method was async was to call the IDocumentMappingService extension method, GetPositionInfoAsync. However, the reason *that* extension method is async is to call GetCodeDocumentAsync(). Ultimately, this method can be made synchronous by passing RazorCodeDocument to it.

This method is async just to call DocumentContext.GetCodeDocumentAsync and the only caller has a RazorCodeDocument in hand. So, this method can be removed in favor of the synchronous version.

This change audits and refactors to the GetCompletionListAsync methods on CompletionListProvider, RazorCompletionListProvider and DelegatedCompletionListProvider in favor of removing async state machines where possible.

A lot of the code to compute completion lists is asynchronous in order
to call `DocumentContext.GetCodeDocumentAsync(...)`. However, many of
these calls can be made synchronous if the `RazorCodeDocument` is
computed up front and passed in. In addition, I've updated some methods
that have both synchronous and asynchronous paths to return `ValueTask`.

I hope that I don't conflict too badly with @davidwengier's completion
work.

CI Build:
https://dev.azure.com/dnceng/internal/_build/results?buildId=2662887&view=results
Test Insertion:
https://dev.azure.com/devdiv/DevDiv/_git/VS/pullrequest/619102