ci: intermittent quality CI failure — ollama model pull connection reset

[planetf1]

Problem

The code-checks / quality CI job occasionally fails at the Pull models step with:

Error: pull model manifest: Get "https://registry.ollama.ai/v2/library/granite4.1/manifests/3b": read tcp ...: read: connection reset by peer

This is a transient network error from Ollama's registry CDN. A single failed pull fails the entire quality run across all three Python matrix versions, requiring a manual re-run.

Observed behaviour

• Failure is transient — a manual re-run of the same commit passes cleanly
• Occurs at the ollama pull granite4.1:3b step in .github/workflows/quality.yml
• The current step has no retry logic, so one connection reset = one wasted run

Fix

Wrap the ollama pull call in a retry loop (5 attempts, 20s backoff, ~2 min total headroom). A PR is linked.