feat: Support vLLM v0.15.1 or newer

[shun-cohere]

Description

This PR aims to support vLLM v0.15.1 or newer verions.
To do this, we introduce a conditional import logic at the top of parser.py

Related Issue

vllm-project/vllm#32240 introduces new structure, which is a breaking change for melody

Motivation and Context

Melody does not support vLLM v0.15 or newer verions.

How Has This Been Tested?

Check1: Confirm that import works with both vLLM versions

$ uv pip list | grep vllm
vllm                              0.15.1
$ uv run cohere_melody_vllm/parser.py
# no error

$ uv pip list | grep vllm
vllm                              0.14.1
$ uv run cohere_melody_vllm/parser.py
# no error

Check2: Tool Calling works with vLLM v0.15.1

Start server

uv run vllm serve CohereLabs/c4ai-command-r7b-12-2024 --reasoning-parser cohere2 --reasoning-parser-plugin ./cohere_melody_vllm/parser.py --tool-parser-plugin ./cohere_melody_vllm/parser.py --tool-call-parser cohere2 --enable-auto-tool-choice

and then send a tool calling query

$ uv run tool.py
ChatCompletion(id='chatcmpl-a8a2b2e52a4dc558', choices=[Choice(finish_reason='tool_calls', index=0, logprobs=None, message=ChatCompletionMessage(content=None, refusal=None, role='assistant', annotations=None, audio=None, function_call=None, tool_calls=[ChatCompletionMessageFunctionToolCall(id='chatcmpl-tool-a9f3c1ba457d277f', function=Function(arguments='{"location": "San Francisco, California", "unit": "celsius"}', name='get_weather'), type='function')], reasoning='I will use the get_weather tool to find out the weather in San Francisco, California in Celsius.', reasoning_content='I will use the get_weather tool to find out the weather in San Francisco, California in Celsius.'), stop_reason=None, token_ids=None)], created=1776133545, model='CohereLabs/c4ai-command-r7b-12-2024', object='chat.completion', service_tier=None, system_fingerprint=None, usage=CompletionUsage(completion_tokens=69, prompt_tokens=1302, total_tokens=1371, completion_tokens_details=None, prompt_tokens_details=None), prompt_logprobs=None, prompt_token_ids=None, kv_transfer_params=None)
'Function called: get_weather'
'Arguments: {"location": "San Francisco, California", "unit": "celsius"}'
'Result: Getting the weather for San Francisco, California in celsius...'

---

Note

Medium Risk
Adds version-dependent imports for vLLM OpenAI protocol types, so a mistake in version detection or module paths could cause runtime import failures across supported vLLM versions.

Overview
Adds vLLM version-aware import logic in cohere_melody_vllm/parser.py to handle the OpenAI entrypoint protocol module reorganization introduced after vLLM 0.14.1, enabling the plugin to run against both old and new layouts.

Updates the Python bindings CI py-check job to run ty check in a matrix against vLLM 0.14.1 and 0.15.1 to continuously validate compatibility.

^{Reviewed by Cursor Bugbot for commit e3aa4f8. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 9e65a48. Configure here.}

[shun-cohere]

To make ty happy, otherwise the following error occurs

error[invalid-method-override]: Invalid override of method `is_reasoning_end`
   --> cohere_melody_vllm/parser.py:144:9
    |
142 |         return content_ids
143 |
144 |     def is_reasoning_end(self, input_ids: list[int]) -> bool:
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Definition is incompatible with `ReasoningParser.is_reasoning_end`
145 |         end_token_id = self.model_tokenizer.convert_tokens_to_ids("<|END_THINKING|>")
146 |         return any(input_id == end_token_id for input_id in reversed(input_ids))
    |
   ::: .venv/lib/python3.12/site-packages/vllm/reasoning/abs_reasoning_parsers.py:54:9
    |
 53 |     @abstractmethod
 54 |     def is_reasoning_end(self, input_ids: Sequence[int]) -> bool:
    |         -------------------------------------------------------- `ReasoningParser.is_reasoning_end` defined here
 55 |         """
 56 |         Check if the reasoning content ends in the input_ids.
    |