Migrate off GPT-4.1 mini before deprecation and make the model configurable per deployment

## Summary
GPT-4.1 mini is being deprecated and needs to be replaced with a newer, supported model in ContentFlow before the retirement date to avoid service disruption. In addition, we propose making the model **configurable per subscription** so future model swaps don't require code changes.

## Background
ContentFlow currently depends on a hardcoded **GPT-4.1 mini** for its model calls. Since this model is scheduled for deprecation, we need to migrate to a supported successor. This was confirmed as still open with Nadeem Ishqair (owner), who requested this issue be filed to track the work.

## Problem
- GPT-4.1 mini will be retired; once deprecated, requests to it will fail.
- ContentFlow will break wherever it references the deprecated model unless migrated in time.
- The model is currently not configurable, so every deprecation or model change forces a code/deployment change. This is a recurring maintenance burden.

## Proposed work

### 1. Immediate migration
1. **Identify all references** to GPT-4.1 mini in ContentFlow (config, deployment names, environment variables, code, prompt/model routing).
2. **Select the replacement model** based on cost, latency, and quality parity for ContentFlow's use cases.
3. **Update configuration/deployment** to point to the new model.
4. **Validate** output quality and regression-test the affected ContentFlow flows.

### 2. Make the model configurable per deployment
- Introduce a **configurable model setting scoped to each deployment** (e.g., a deployment-level config value / environment variable) instead of a hardcoded model reference.
- ContentFlow should read the model name/deployment from this configuration at runtime, with a sensible **default** and validation/fallback if an unsupported value is set.
- This lets each deployment select its own model and lets us swap models on future deprecations **without code changes**.

Start by searching ContentFlow for GPT-4.1 mini references across configuration, deployment names, environment variables, code, and prompt/model routing. Review the current model-call entry points and affected flows, then define the replacement and deployment-scoped setting. Done means a supported model is used, each deployment can configure it with a default and validation/fallback, and affected flows pass regression checks.

Written by the indexing model from the issue text.