AnswerRelevancy uses model parameter instead of embedding_model for Embedding call

[cgoer]

Hi there,
just found a possible bug. The AnswerRelevancy scorer allows to pass an embedding_model parameter, which is used in the EmbeddingSimilarity call of the async runner, but not in the synchronous version, leading to wrong model usage when using the scorer synchronously.

EmbeddingSimilarity(client=self.client).eval_async(
    output=q["question"], expected=input, model=self.embedding_model
)

vs.

EmbeddingSimilarity(client=self.client).eval(output=q["question"], expected=input, model=self.model)

created a PR: #167