fix: support max_completion_tokens for reasoning models

[tink-bot]

Reasoning models (o1/o3/o4/gpt-5 series) unusable with this action — go-openai's ReasoningValidator rejects requests client-side: max_tokens set = this model is not supported MaxTokens, please use MaxCompletionTokens; temperature other than 1 fails too. Action always sends both (defaults 1000 / 0.7), so every reasoning-model run errors before any HTTP request.

Fixes #23

Changes:

• New max_completion_tokens input, takes precedence over max_tokens.
• Auto-remap: reasoning model + only max_tokens set = budget sent as max_completion_tokens, existing configs keep working without changes.
• Temperature omitted for reasoning models (fixed at 1 by the API); note printed when a different value was configured. Non-reasoning models unchanged.
• Tests for config parsing and request building, README + action.yml docs updated.

How to verify

1. Run the action with model: gpt-5.6-luna (or any o1/o3/o4/gpt-5 model), max_tokens: '4000', and temperature: '0.2' against the OpenAI API.
2. Expected: The request succeeds, the log contains a note that the temperature value is omitted, and the response is generated with a completion budget of 4000 tokens.
3. Run the action with a non-reasoning model such as gpt-4o and the same inputs.
4. Expected: The request behaves exactly as before this change, with max_tokens and temperature sent as configured.

Before this PR: step 1 failed with chat completion error: this model is not supported MaxTokens, please use MaxCompletionTokens.

[kolaente]

This PR is entirely AI generated and reviewed by me, feel free to discard it completely