chore: normalize CSVs to LF + enforce in CI

[pierreleripoll]

Normalize every tracked CSV to LF line endings with a final newline, and enforce it in CI.

Why

The hub's contribution-submit feature (see resslab-hub-data#41) re-serializes the primary CSV with LF endings when appending contributed rows. Five of the nine primary CSVs were stored with CRLF, so every contribution PR showed the entire file — header included — as changed, making diffs unreviewable and guaranteeing conflicts between concurrent contributions.

What

• chore: normalize all CSVs to LF line endings with final newlines — 1,167 files: 1,165 CRLF→LF, 159 missing final newlines added (157 overlap). Verified: git diff --ignore-cr-at-eol against the base is completely empty, i.e. zero content changes; all 9 primary CSVs now round-trip byte-identically through the backend's submission serializer, so future contribution diffs are exactly the appended rows.
• ci: enforce LF + final newline on CSVs — .gitattributes gains *.csv text eol=lf (git clients auto-normalize on commit), and data-validation.yml gains an unconditional line-endings job (fails on any CR in a tracked CSV or any missing final newline) as the backstop for web uploads and API commits, which bypass .gitattributes.

Merging notes

• Squash-merge with [skip ci] in the commit title — e.g. chore: normalize CSVs to LF + enforce in CI [skip ci]. Otherwise ci.yml re-uploads ~1,165 unchanged files to the prod database.
• The PR-validation jobs for all 6 mapped types will run on this PR; content is unchanged so they should pass as they do today.
• Open data PRs (including data(material): add 15 experiments (DEMO-MAT-001…) #41) merge cleanly afterwards: both sides make identical line-ending changes, which git auto-resolves.
• Follow-up after merge: add .git-blame-ignore-revs containing the squashed commit's SHA (it can't be known before the squash), so GitHub blame and git blame skip this commit.