Update scPRINT checkpoint names

[jkobject]

Summary

• update scPRINT variants and defaults to checkpoint filenames currently published in jkobject/scPRINT
• keep the script fallback default aligned with medium-v1.5
• retain only the two checkpoint-name commits on top of current main; historical merged/scGPT changes were removed

Relation to #87

This is the maintained scPRINT-only successor to #87. The scGPT deletions from #87 are intentionally not carried forward: reviewer feedback correctly notes that setup: must remain, and #87 also displayed the already-merged max_len change from #86 because of historical branch ancestry.

Validation

• current Viash reusable workflow is running on the rebased immutable head
• YAML model-name contract parsed successfully
• Python script compiles
• large-v1.ckpt, medium-v1.5.ckpt, and small-v1.ckpt verified in the Hugging Face model repository

Supersedes #87; please do not merge both.

[jkobject]

Independent review: PASS

Reviewed exact head ebc7bf2b8d0fa3ef54d75f4946629aa06dfcb1bf against current base 68eb2ecc52990617dbb6d1bb5c7158d9893796bb.

Blocking findings: none.

Evidence:

• The base is the first parent of the current clean merge ref; the PR is exactly 2 commits ahead and 0 behind.
• The diff is one-purpose: only src/methods/scprint/config.vsh.yaml (6+/6-) and src/methods/scprint/script.py (1+/1-). It updates checkpoint identifiers consistently to large-v1, medium-v1.5, and small-v1; it contains no scGPT edits and no max_len change.
• Fresh CI run 31571732777 is SUCCESS at this exact head. Changed-component detection selected only methods/scprint; the scPRINT job built the component, downloaded small-v1.ckpt, produced and validated the output, and ended SUCCESS! All 2 out of 2 test scripts succeeded!.
• Independent static checks: YAML parses; variants, choices, default, and test model agree; script.py compiles; git diff --check passes. Direct HTTP readback returned 200 for all three named checkpoint files.
• No review threads or unresolved comments exist on this PR.

Verdict: maintainer-ready. This is a review comment rather than a GitHub approval because the reviewer is operating through the PR author's account.

[rcannood]

Thank you @jkobject !