fix(ci): use llama3.2:1b in CI, increase Ollama timeout, add --model arg

[JerrettDavis]

Root Cause

Step 8 'Synthesize training examples' was failing with TaskCanceledException in GitHub Actions CI due to:

• Wrong model selection: qwen3.5:9b (6.6GB) too large/slow for CPU
• HttpClient timeout too short (300s) for slow CPU inference

Changes

• train-model.yml: Switch CI model from qwen3.5:9b → llama3.2:1b (1.3GB, much faster); reduce default examples 150→50; pass --model arg to synthesize and validate steps
• OllamaClient.cs: Increase HttpClient.Timeout from 5min → 10min to handle slow inference
• Program.cs: Add --model CLI arg (default qwen3.5:9b) forwarded to AiTrainingDataSynthesizer for runtime override

Fixes

Resolves workflow run #24970871185 TaskCanceledException

[github-actions]

⚠️ Deprecation Warning: The deny-licenses option is deprecated for possible removal in the next major release. For more information, see issue 997.

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None

[github-actions]

Code Coverage

Summary
  Generated on: 04/27/2026 - 01:11:04
  Coverage date: 04/27/2026 - 01:09:03 - 04/27/2026 - 01:10:56
  Parser: MultiReport (7x Cobertura)
  Assemblies: 23
  Classes: 1067
  Files: 591
  Line coverage: 74.6%
  Covered lines: 36802
  Uncovered lines: 12510
  Coverable lines: 49312
  Total lines: 97270
  Branch coverage: 63.4% (14287 of 22512)
  Covered branches: 14287
  Total branches: 22512
  Method coverage: 86.3% (6308 of 7308)
  Full method coverage: 74% (5415 of 7308)
  Covered methods: 6308
  Fully covered methods: 5415
  Total methods: 7308