[BUG] LiteLLM provider sends max_tokens instead of max_completion_tokens for models that require the latter (400 BadRequest)

[TGDone]

Version: 3.85.100536
Provider: LiteLLM (OpenAI-compatible proxy)
Model: gpt-6-sol (routed to an OpenAI model)

Steps to reproduce

1. Configure the LiteLLM provider with a model that only accepts max_completion_tokens.
2. Send any message (e.g. "what model are you?") in Ask mode.

Expected
The request succeeds.

Actual
400 error:
OpenAIException - Unsupported parameter: 'max_tokens' is not supported
with this model. Use 'max_completion_tokens' instead.

Analysis
The extension appears to always send max_tokens in requests to the LiteLLM
provider. LiteLLM cannot infer that this custom model alias is a reasoning /
new-generation model, so it does not convert the parameter.

Suggested fix

• Send max_completion_tokens for LiteLLM, or
• Add a setting to choose which parameter name is used (or omit it).

Logs
(redacted error JSON attached)