This PR changes the IL generated forstackalloc in the following way:

• If an existing optimization applies (loadingbyte arrays directly from PE data; usinginitblk when all bytes are the same), there is no change.
• Otherwise, if it's converted toReadOnlySpan<T> and all initializers are constant, the wholestackalloc is replaced by a call toCreateSpan.
• Otherwise (e.g. when converted toSpan<T> or when only some initializers are constant),CreateSpan is used as a part of regularstackalloc codegen, together withcpblk.

The IL for array initializers is also changed:

• If an array initializer for a type larger thanbyte with all initializers constant is converted to ROS, the whole expression is replaced withCreateSpan.

There are still some questions I have and things that would need addressing to make this code production quality:

1. Is it okay to callGetPinnableReference (without any pinning) to get the backing data for the result ofCreateSpan, for use incpblk?
2. Alignment. As I understand it, the compiler currently does not align the metadata blocks containing initialization data in any way. If this is a requirement, that would need to change.
3. ENC. This optimization does not work with ENC and the code needs to know whether to apply this optimization at the lowering stage. But I don't know how to find out if ENC is used from inside ofLocalRewriter. Is there some way to do that?