Some behavioral / perf fixes for Embedding<T>

[stephentoub]

Motivation and Context

Clean up some Embedding<T> behaviors / perf issues.

Description

• IsSupported was more expensive than it needs to be: if it's changed to just compare the types directly rather than searching a list, the JIT can turn the entire operation into a JIT-time constant.
• The explicit cast to ReadOnlySpan<T> was cloning the array. It doesn't need to do that.
• default(Embedding<T>) results in a struct that fails on a bunch of operations with NullReferenceExceptions if you try to use it. We've learned from experience with ImmutableArray<T> this is suboptimal. I've made it functional and behave equivalent to Empty... in fact, Empty is now just default.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows SK Contribution Guidelines (https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md)
• The code follows the .NET coding conventions (https://learn.microsoft.com/dotnet/csharp/fundamentals/coding-style/coding-conventions) verified with dotnet format
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

[MovGP0]

I don't like code in the form of private readonly TEmbedding[]? _vector; as seen in Embeddings.cs.

Enumerable objects should return an empty enumerable object like Array.Empty<T> when not initialized instead of the null value.

[stephentoub]

Enumerable objects should return an empty enumerable object like Array.Empty when not initialized instead of the null value.

Where is it returning null?