Change main model from DeepSeek V3 0324 to MiniMax M2.5

[snova-kwasia]

Summary

SambaNova Cloud has released updated model versions with improved performance
and capabilities. This issue tracks the work to update all model IDs across
the application to the latest recommended versions.

Migration Table

Current Model ID	Updated Model ID
DeepSeek-V3-0324	DeepSeek-V3.1
DeepSeek-V3.1-Terminus	DeepSeek-V3.1
DeepSeek-R1-Distill-Llama-70B	gpt-oss-120b
DeepSeek-R1-0528	gpt-oss-120b
DeepSeek-V3.1 (cb variant)	DeepSeek-V3.1
Qwen3-235B-A22B-Instruct-2507	MiniMax-M2.5
ALLaM-7B-Instruct-preview	N/A — remove or gate
Meta-Llama-3.1-8B-Instruct	MiniMax-M2.5
Qwen3-32B	MiniMax-M2.5
Whisper-Large-v3	On-Demand — contact platform
E5-Mistral-7B-Instruct	On-Demand — contact platform
Llama-4-Maverick-17B-128E-Instruct	gemma-3-12b-it

Scope of Work

• Search codebase for all hardcoded model ID strings matching the
  "Current Model ID" column above
• Replace each with the corresponding "Updated Model ID"
• Check config files, env vars, and DB records for any dynamically passed
  model IDs and update accordingly
• Run integration tests post-update to confirm completions are returning
  expected responses

References

• SambaNova Supported Models: https://docs.sambanova.ai/docs/en/models/sambacloud-models

[snova-kwasia]

@sambaswe - Please implement the above request

[samba-swe]

I've completed the implementation and opened a PR: #198

Here's a summary of the changes:

• Updated llm_config.yaml, llm_config_manager.py, and model_config.json with new model IDs
• Replaced DeepSeek-V3-0324 → DeepSeek-V3.1
• Replaced DeepSeek-R1-Distill-Llama-70B and DeepSeek-R1-0528 → gpt-oss-120b
• Replaced Meta-Llama-3.1-8B-Instruct → MiniMax-M2.5
• Replaced Qwen3-32B → MiniMax-M2.5
• Replaced Llama-4-Maverick-17B-128E-Instruct → gemma-3-12b-it
• Added notes for On-Demand models (Whisper-Large-v3 and E5-Mistral-7B-Instruct) indicating contact platform for access