fix(openai_compatible): handle None tool.description in function_call format and token counting

PromptMessageTool.description is typed as str but can be None in
practice when using custom OpenAPI-based tool providers. This causes
two crashes:

1. Line 601: Passing None as 'description' in the function_call
   payload sent to the LLM API, which may reject it.

2. Line 1130: _get_num_tokens_by_gpt2(tool.description) crashes with
   TypeError when description is None, since gpt2 tokenizer expects
   a string.

Both are fixed with 'tool.description or ""' — a minimal, safe
fallback consistent with the pattern already used in the azure_openai
plugin (line 472).

Reproduction: Create an OpenAPI tool provider whose endpoints have a
summary but no description, then invoke an agent using those tools.

Related: langgenius/dify-official-plugins#3095 (same bug in the
azure_openai plugin's own _num_tokens_for_tools)

Author: Javier Lázaro

src/dify_plugin/interfaces/model/openai_compatible/llm.py

@@ -598,7 +598,7 @@ def _generate(
                 data["functions"] = [
                     {
                         "name": tool.name,
-                        "description": tool.description,
+                        "description": tool.description or "",
                         "parameters": tool.parameters,
                     }
                     for tool in tools
@@ -1127,7 +1127,7 @@ def _num_tokens_for_tools(self, tools: list[PromptMessageTool]) -> int:
                 num_tokens += self._get_num_tokens_by_gpt2(tool.name)
             num_tokens += self._get_num_tokens_by_gpt2("description")
             if hasattr(tool, "description"):
-                num_tokens += self._get_num_tokens_by_gpt2(tool.description)
+                num_tokens += self._get_num_tokens_by_gpt2(tool.description or "")
             if hasattr(tool, "parameters"):
                 parameters = tool.parameters
                 num_tokens += self._get_num_tokens_by_gpt2("parameters")