fix(multimodal): honor effective CPU/GPU config for mmproj

[leehack]

Summary

• align multimodal projector offload with effective model-load configuration so CPU-only paths keep mmproj on CPU
• persist per-model mmproj offload intent after backend fallback and apply it when creating the multimodal context
• add unit coverage for resolver behavior and document the CPU-only projector behavior in docs/changelog

Fixes #63

Testing

• dart analyze lib/src/backends/llama_cpp/llama_cpp_service.dart test/unit/backends/llama_cpp/llama_cpp_service_test.dart
• dart test test/unit/backends/llama_cpp/llama_cpp_service_test.dart
• dart test test/unit/core/engine/engine_test.dart

[codecov-commenter]

Codecov Report

❌ Patch coverage is 87.50000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 76.51%. Comparing base (ebaec0b) to head (c11cd1b).

Files with missing lines	Patch %	Lines
lib/src/backends/llama_cpp/llama_cpp_service.dart	87.50%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #69      +/-   ##
==========================================
+ Coverage   76.50%   76.51%   +0.01%     
==========================================
  Files          66       66              
  Lines        8188     8196       +8     
==========================================
+ Hits         6264     6271       +7     
- Misses       1924     1925       +1     

Flag	Coverage Δ
unittests	76.51% <87.50%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.