fix(litellm): explicitly pass encoding_format="float" (#144)

LiteLLM documents encoding_format as defaulting to float, but in
practice some providers (e.g. nvidia_nim) error out when it's not
provided. Pass it explicitly as a workaround.