[Enhancement]: Add native Mistral provider support

Target Component

External Integrations (LLM/Search APIs)

Enhancement Description

PentAGI has no native Mistral provider, so Mistral users must configure it through the generic custom provider (LLM_SERVER_URL=https://api.mistral.ai/v1). That path does not work reliably with Mistral's hosted reasoning models, including the new hosted GLM 5.2/5.3 models.

PentAGI sends the modern reasoning: {effort: ...} object, which Mistral rejects as an unknown field (extra_forbidden: body.reasoning). The warning log shows only an empty body (API returned unexpected status code: 422: ), hiding the root cause.

Mistral-hosted models (zai-glm-latest, zai-glm-5-2, zai-glm-5-3, magistral) return the assistant message as an OpenAI-style content-block array instead of a plain string:

"content": [
    {"type": "thinking", "thinking": [{"type": "text", "text": "..."}]},
    {"type": "text", "text": "..."}
]

This kills every agent chain with:

error decoding response: json: cannot unmarshal array into Go struct field ChatCompletionChoice.choices.message.content of type string

LLM_SERVER_PRESERVE_REASONING=true does not help: it only affects request-side history re-feeding, not response parsing.

zai-glm-latest (5.3) emits thinking blocks even at reasoning_effort: low (non-deterministically on complex prompts), and Mistral only accepts low/high/max, so that model is unusable through the custom provider. zai-glm-5-2 works only with no reasoning parameter at all, meaning no reasoning.

The goal: a first-class Mistral provider that lets PentAGI drive Mistral's hosted GLM 5.2/5.3 (and the rest of its reasoning catalog) with full reasoning support, from a single Mistral API key.

Technical Details

I can propose a PR if this sounds interesting.

Designs and Mockups

No response

Alternative Solutions

No response

Verification

• I have checked that this enhancement hasn't been already proposed
• This enhancement aligns with PentAGI's goal of autonomous penetration testing
• I have considered the security implications of this enhancement
• I have provided clear use cases and benefits