feat: further vram optimizations

[avinash2692]

Further optimizations to the GPU VRAM when running tests.

Type of PR

• Bug Fix
• New Feature
• Documentation
• Other

Description

###F ix GPU OOM in test runs: ollama lifecycle ownership + VRAM timing**

• run_tests_with_ollama.sh starts and warms up all three ollama models unconditionally, even when an external process has already started the server — fragmenting VRAM before HF tests run and triggering OOM
• Add OLLAMA_EXTERNAL and OLLAMA_SKIP_WARMUP env guards to the shell script so an external orchestrator can own the lifecycle (both default to 0 so standalone use is unaffected)
• conftest.py handles VRAM timing — loading models right before the ollama test group and evicting after via keep_alive
• Fix _check_ollama_available() hardcoded port to respect OLLAMA_HOST/OLLAMA_PORT env vars
• Add extra GPU cleanup in test_alora_train_integration.py for all tests.

Testing

• Tests added to the respective file if code was changed
• New code has 100% coverage if code as added
• Ensure existing tests and github automation passes (a maintainer will kick off the github automation when the rest of the PR is populated)

[github-actions]

The PR description has been updated. Please fill out the template for your PR to be reviewed.