Add Faraday retry support for the OpenAI LLM

[SpiffyStores]

Currently there is no support for retrying OpenAI requests that fail due to network glitches or other reasons.

When processing a very large number of embed! requests, the process can crash fairly often due to temporary network issues. This means that the updates to the vector store may be lost as it's difficult to determine where the failure occurred. Re-running a very large number of embedding requests is often not very practical as it can sometimes take days to process everything.

There's a very simple fix available using the Faraday retry middleware.

Here's the initialization method with retry middleware added.

retry_options = {
  max: 6,
  interval: 0.05,
  interval_randomness: 0.5,
  backoff_factor: 2,
  retry_statuses: [429, 500, 502, 503, 504],
  methods: %i[get post],
  exceptions: Faraday::Retry::Middleware::DEFAULT_EXCEPTIONS + [Faraday::ConnectionFailed]
}

@client = ::OpenAI::Client.new(access_token: api_key, **llm_options) do |f|
  f.response :logger, Langchain.logger, {headers: true, bodies: true, errors: true}
  f.request :retry, retry_options
end

So far, I've found these parameters to be fine with the OpenAI API.