Add analyzer/fixer to recommend updating stackalloc expressions to use collection expressions.

[CyrusNajmabadi]

Part of #69132.

Followup to #69407. That should be reviewed and go in first.

Supports cases like:

ReadOnlySpan<int> x = stackalloc int[] { 1, 2, 3 };

// And converts it to:
ReadOnlySpan<int> x = [1, 2, 3];

Also supports;

ReadOnlySpan<int> x = stackalloc int[3];
x[0] = a;
x[1] = b;
x[2] = c;

// And converts it to
ReadOnlySpan<int> x = [a, b, c];

[CyrusNajmabadi]

@akhera99 @genlu @Cosifne this is ready for review now that #69407 has gone in.

[CyrusNajmabadi]

review with whitespace off.

this is just a method inline.

[CyrusNajmabadi]

this is just a method inline.

[CyrusNajmabadi]

this code is saying: it's ok if using a collection expression would produce Span<T> if the original code was ReadONlySpan<T>. That's because the former is trivially implicitly convertible to the latter, despite being different types.

[CyrusNajmabadi]

this moved from common code to be specific to collection-expressions and specific to colleciton-initializers. it turned out sharing just was not possible. Especially because if you had x[a] = b then collection expressions would add this as just b whereas collection initializers would add it as [a] = b.

Given that, it would be too many contortions to try to share. so i just pulled out for eahc.

[CyrusNajmabadi]

we want different equivalence keys so that we don't wrong the wrong fixers on the wrong diagnostics.

[CyrusNajmabadi]

this is an improvement. the latter is legal in a collection expression (but was not legal in a collection initializer). No longer sharing code between the two improved this.

[CyrusNajmabadi]

swithcing to SemanticDOcument moved us from 4 local variables to 1.

[CyrusNajmabadi]

only one variable to update versus 4.