Add ChecksumAlgorithm to project snapshot

[tmat]

The goal is to allow EnC analyzer to check whether content of a Document matches the checksum stored in the PDB when applying EnC/Hot Reload changes. Currently we read the content from disk since we can't get the correct checksum from Document and such approach runs into issues when the file is updated on disk before we can read its content. This results in EnC/Hot Reload incorrectly blocking or ignoring changes.

Adds ChecksumAlgorithm to Project snapshot (on ProjectAttributes). The value is derived from the corresponding msbuild property set in project file.

Checksum algorithm of a Document is stored on DocumentState and initialized from the TextLoader when the loader is set to the state. This is needed because the state does not directly hold on the loader, but instead it holds on a value source for the SourceText or SyntaxTree. It's not part of DocumentAttributes as we would need to keep the attributes in sync with the value of the loader.

Fixes #51993
Fixes AB#1567043
Implements #64220

Bans use of APIs in Roslyn whose implementations create SourceText but do not have a checksum algorithm parameter and always use the default (SHA1). In Roslyn product code we should be explicit about what checksum algorithm is used.

Allow usage of APIs listed in eng\config\BannedSymbols.txt in tests without warnings. These APIs would ideally not be used at all but usage in tests doesn't matter much. In fact, we still need to test these APIs even though we don't want them used in product code. Avoiding usage of APIs banned in this PR in all tests is not worth the significant effort it would take to clean up all tests.

This PR keeps new TextLoader API internal. These will become public in the follow up change: #64192

PdbMatchingSourceTextProvider (96e0f77)

EnC needs to ensure for each modified document that it has a baseline snapshot of that document whose content exactly matches the source used for compilation of the baseline binary. We use checksum stored in the PDB for the comparison.

When debugging session starts, we stash away the current Solution snapshot so that we can compare updated documents against it. However, this snapshot doesn't necessarily have all file content loaded at this time. By the time we need to compare the content the file content on disk might have changed. To mitigate this issue, we pre-loaded all open documents at the start of debugging session and then watched the workspace for new documents being opened. When the document was opened, we notified the EnC service to check if it already has the PDB matching content snapshot for this document. If not, it would try to read the current file content and see if it's matching the PDB.

This approach has an inherent race condition. This PR changes implements a different approach. We record document snapshots of interest (candidates for PDB matching content) in reaction to workspace events. Instead of notifying the EnC service at that point we let it to call back to the workspace listener when it encounters a document whose baseline it doesn't have yet.

VS validation 1: https://dev.azure.com/devdiv/DevDiv/_git/VS/pullrequest/426348
VS validation 2: https://devdiv.visualstudio.com/DevDiv/_build/results?buildId=6790099&view=results

[tmat]

@jasonmalinowski @dotnet/roslyn-compiler PTAL

[tmat]

Found some issues and redoing the representation of checksum alg in Document snapshot.

[tmat]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 4 pipeline(s).