change uses of Embedding.weight to Embedding.asLinear

[davidkoski]

There are several Modules that use

open class Embedding: Module, UnaryLayer, Quantizable {
    public let weight: MLXArray

directly instead of using:

    /// Call the embedding layer as a linear layer.
    ///
    /// Use this for example when input embedding and output projection
    /// weights are tied.
    open func asLinear(_ x: MLXArray) -> MLXArray {
        matmul(x, weight.T)
    }

e.g. Cohere, Starcoder2, OpenELM (this last is fixed recently).

        out = matmul(out, model.embedTokens.weight.T)

should be:

        out = model.embedTokens.asLinear(out)