chore: no .gitattributes across the org, so line endings are unmanaged

[BryanFRD]

Tracking issue for a one-line-per-repo fix. Referenced from the PRs rather than duplicated as twenty near-identical issues.

The problem

Most repos in the org have no .gitattributes. With core.autocrlf enabled, a Windows checkout gets CRLF in the working tree while every formatter in the toolchain (prettier, rustfmt, gofmt) writes LF. The result is a permanent, meaningless diff between what is on disk and what the tools produce.

Measured on FerrGames-Cloud: 19 files failed prettier --check and running prettier --write over them produced an empty git diff. Not one had a real formatting problem. The consequence is that nobody trusts the check, so nobody runs it, so real drift accumulates unnoticed and no formatting gate can be added to CI.

What is actually stored

Nothing needs rewriting. Across all 19 checkouts surveyed, zero text files are stored with CRLF in the index; git already normalises on commit. This is purely about what a checkout puts on disk, which is why each fix is a single new file and git add --renormalize . stages nothing.

The fix

* text=auto eol=lf, binaries declared explicitly so their bytes are never inspected, and .bat / .cmd kept at CRLF for cmd.exe.

FerrLabs/ia already carries exactly this rule, with the same reasoning in its comment, so this aligns the rest of the org with an existing precedent rather than inventing one.

Scope

New file (17): Benchmarks, Changelog, Docker-Runners, FerrFleet-Cloud, FerrFlow-Cloud, FerrGames-Discord, FerrGrowth-Cloud, FerrLabs-Cloud, .github, FerrLens-Cloud, FerrTrack-Cloud, FerrVault-Cloud, Infra, Kit, MCP, Status, UI.

Extend an existing partial file (3): FerrFlow, Fixtures and FerrVault (operator) each declare .githooks/* or *.sh only, and so carry the same problem for everything else.

Already correct (1): ia.

FerrGames-Cloud landed first, as FerrLabs/FerrGames-Cloud#1050.

Note for contributors

The first checkout after each merge rewrites line endings across that working tree once. It is local churn and changes no content.

[BryanFRD]

Review feedback across the batch, recorded here rather than spread over twenty PRs.

Every PR was approved with no blocking issue. Several reviewers independently raised the same nit: the explicit binary list covers images, fonts, audio, video and archives, but not .eot, .jar, .exe, .dll, .so. Each of them also checked their own repo and confirmed none of those files exist there today.

The nit is fair on its own terms. The reason the list exists at all is that I did not want to trust text=auto's NUL-byte heuristic with images, and that argument applies to executables just as well.

Not acting on it now, deliberately. Extending the list means either twenty more PRs for a file type no repo currently has, or patching only the PRs still open and leaving the org with two different versions of the same file. A uniform list that is slightly short beats a divergent one, and every format named above carries NUL bytes in its first few hundred, so the heuristic covers them in practice.

Worth doing as a single sweep the day the first such file actually lands, along with anything else that has shown up by then.