Add first-class LLMGateway.io provider

[goianiense]

Summary

• Add a dedicated :llm_gateway provider (modeled on OpenRouter) for LLMGateway.io, an OpenAI-compatible gateway for 35+ upstream providers.
• Register config options llm_gateway_api_key / llm_gateway_api_base (default https://api.llmgateway.io/v1) so the gateway can be used alongside the real OpenAI provider instead of hijacking openai_api_base.
• Include chat completions, models catalog parsing, streaming, image generation, and files support, with unit coverage for config, API base resolution, and error parsing.

Closes #821.

Usage

RubyLLM.configure do |config|
  config.openai_api_key = ENV["OPENAI_API_KEY"]
  config.llm_gateway_api_key = ENV["LLM_GATEWAY_API_KEY"]
end

# Gateway and OpenAI work independently
RubyLLM.chat(model: "gpt-4o", provider: :llm_gateway)
RubyLLM.chat(model: "gpt-4o", provider: :openai)

Test plan

• bundle exec rspec spec/ruby_llm/providers/llm_gateway/parse_error_spec.rb spec/ruby_llm/provider_spec.rb spec/ruby_llm/configuration_spec.rb (25 examples, 0 failures)
• Smoke test: separate API bases/keys for :llm_gateway and :openai
• Optional: live chat against LLMGateway.io with LLM_GATEWAY_API_KEY set