* Drop previous_response_id for Conversations API

OpenAI's Conversations API doesn't support forking - conversation_id
and previous_response_id are mutually exclusive parameters. Since the
gem always creates conversations, previous_response_id was never being
used effectively.

This simplifies the gem to focus solely on the Conversations API
pattern where conversation_id maintains context across turns.

* Cleanup and simplification (#46)

Default configuration:
- Model: gpt-5.1 (was gpt-4.1-nano)
- reasoning_effort: nil (equivalent to "none" for gpt-5.1)
- All tools (web_search, code_interpreter, image_generation) now
  work out of the box without configuration

Code changes:
- Add last_response_id as public reader for background mode workflows
- Change web_search_preview tool to web_search (matches OpenAI GA)
- Use last_response_id accessor internally instead of @instance_var

Test improvements:
- Add tests for conversation continuity across generate! calls
- Add tests for continuing conversations in new instances
- Add tests for file handling and conversation items retrieval
- Add test for last_response_id updates
- Remove low-value and redundant tests

Documentation:
- Update CHANGELOG.md with 0.4.0 changes
- Update DEVELOPMENT_JOURNEY.md helper methods section
- Simplify README by removing tool compatibility warnings
- Add note that nil reasoning equals "none" for gpt-5.1

Examples:
- Renumber example files (now 01-15)
- Remove redundant/low-value examples (03_configuration,
  08_advanced_usage, tmp_image)
- Remove reasoning_effort=nil lines (no longer needed)

Infrastructure:
- Fix Gemfile.lock to use openai gem from RubyGems
- Delete local openai-ruby reference directory