update_gemini_kwargs mutates caller generation config

[lntutor]

Bug

instructor.v2.providers.gemini.utils.update_gemini_kwargs() shallow-copies the request dictionary but mutates the nested generation_config dictionary in place while translating OpenAI-style keys.

Reproduction

from instructor.v2.providers.gemini import utils

utils._default_safety_thresholds = lambda: None
generation_config = {"max_tokens": 5}

utils.update_gemini_kwargs({"generation_config": generation_config})
print(generation_config)

Current output:

{"max_output_tokens": 5}

Expected:

{"max_tokens": 5}

The function should treat caller-owned configuration as read-only and return a normalized copy.

Impact

Reusing the same options dictionary for a retry or later request silently changes the accepted parameter names and can produce behavior that depends on call order.