refactor: align OpenRouter reasoning support with MistralChatGenerator

- Accept `list[ChatMessage] | str` in run/run_async and normalize via
  `_normalize_messages`. Fixes the mypy override (LSP) error and restores the
  string-input support that OpenAIChatGenerator now provides.
- Stop overriding the streaming handlers and the chunk converter; inherit them
  from the parent and instead warn when streaming is combined with reasoning,
  since reasoning content can only be reconstructed from non-streaming
  responses (mirrors MistralChatGenerator). This removes ~150 lines of
  duplicated parent code that also silently dropped `reasoning_details` during
  streaming, so multi-turn reasoning pass-back no longer no-ops when streaming.
- Restore the explanatory comments that were dropped from `to_dict` and
  `_prepare_api_call` (the `openai_endpoint` hint mechanism and the `to_dict`
  override rationale).
- Bump the `haystack-ai` dependency to >=2.30.0 for `_normalize_messages`.
- Tests: add string-input and streaming-with-reasoning-warning tests; drop the
  tests for the removed streaming converter.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: julian-risch

integrations/openrouter/pyproject.toml

@@ -23,7 +23,7 @@ classifiers = [
   "Programming Language :: Python :: Implementation :: CPython",
   "Programming Language :: Python :: Implementation :: PyPy",
 ]
-dependencies = ["haystack-ai>=2.22.0"]
+dependencies = ["haystack-ai>=2.30.0"]
 
 [project.urls]
 Documentation = "https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/openrouter#readme"