Update use of HashAlgorithm to reduce allocations

[sharwell]

Version Used: 15.1

PerfView is showing substantial overhead for the creation of a temporary array in HashAlgorithm.ComputeHash(Stream), originating from Microsoft.CodeAnalysis and Microsoft.CodeAnalysis.Workspaces. On my machine, this accounted for more than 3 seconds of time while Roslyn.sln was opening. Code using this method should be updated to use an alternative that allows the use of a caller-specified buffer from a pool.

[CyrusNajmabadi]

tagging @heejaechang

From talking to heejae earlier, we're nto certain if anything can be helped here. The issue is that we're not computing the hash and throwing away the value. We hold onto the value. So pooling isn't helpful.

[sharwell]

@CyrusNajmabadi The time spent is in clr!JIT_NewArr1. It's literally the buffer allocation using for reading from the stream that's taking time.

[CyrusNajmabadi]

Related: https://github.com/dotnet/corefx/issues/16622

[CyrusNajmabadi]

Can you be more specific sam? I can imagine a lot of different buffers at play here.