feat: add SHA256 checksums for remaining GGUF models

[boffin-dmytro]

Follow-up to PR #169. Adds verified checksums for the 3 models that were left empty.

Changes

• NV_ULTRA (qwen3-coder-next): 9e6032d2f3b50a60f17ce8bf5a1d85c71af9b53b89c7978020ae7c660f29b090
• SH_LARGE (qwen3-coder-next): 9e6032d2f3b50a60f17ce8bf5a1d85c71af9b53b89c7978020ae7c660f29b090
• Tier 3 (Qwen3-14B): 5eaa0870bd81ed3b58a630a271234cfa604e43ffb3a19cd68e54a80dd9d52a66

Completes integrity verification coverage for all available models.

Testing

./tests/test-model-integrity.sh
# Results: 6 passed, 0 failed
# Checksums defined: 7 (up from 4)

Source

Checksums verified and sourced from PR #165 as recommended by @Lightheartdevs in PR #169 review.

Files changed: 1 file, 3 lines modified

[Lightheartdevs]

Review: ✅ APPROVE

Clean, minimal, high-value change that completes the model integrity verification started in PR #169.

What it does: Fills in the three remaining empty GGUF_SHA256="" fields — two for qwen3-coder-next-Q4_K_M.gguf (tiers SH_XLARGE and SH_LARGE, same file so same hash) and one for Qwen3-14B-Q4_K_M.gguf (tier 3).

Verified:

• Both hashes are valid 64-char hex SHA256 format
• The two Strix Halo tiers correctly share the same hash (same GGUF URL) — consistent with the existing pattern where tiers 1/2 and SH_COMPACT/4 share hashes for the same files
• After this PR, every non-CLOUD tier with a GGUF download URL will have a corresponding SHA256 checksum

CI note: All failures are pre-existing (gitleaks license issue for orgs + distro smoke test failures on main). Shell lint passes, which is the relevant check.

Merge when ready.