Add media compression workflow

[MUFFANUJ]

closes #289

This PR adds a new CI workflow to enforce media optimization. On pull requests to main, it compresses changed png/gif/mp4 files, uploads optimized outputs as the compressed-media-assets artifact, and fails the check when compression updates are detected so they can be included in the PR. It also provides a workflow_dispatch trigger to run the same compression check across all tracked media files for one-time repository backfill.

[MUFFANUJ]

This PR adds stronger media compression checks in CI: non-snapshot PNGs are compressed with pngquant (with optipng fallback), *-snapshots/*.png remain lossless to reduce visual-test drift, GIFs are compressed with gifsicle --lossy=120 --colors 128, and MP4s are compressed with ffmpeg (crf=32, fps=20, scaled to fit within 1280x720 while preserving aspect ratio). The workflow uploads optimized artifacts and fails when compressible changes are detected so the optimized outputs can be applied in the PR. It also introduces .github/media-compression.lock (path + SHA-256) to make behavior deterministic: files already matching the lock are skipped, and when CI compresses a file it updates the lock entry and includes the updated lockfile in the artifact, so applying the artifact and pushing prevents repeat fail loops on the same files. On the same ~16.42 MB media set, research runs showed conservative lossless compression to ~15.08 MB (save ~1.34 MB, 8.16%), moderate lossy to ~11.95 MB (save ~4.47 MB, 27.25%), and aggressive lossy to ~9.7–9.8 MB (save ~6.6–6.7 MB, ~40%). This workflow uses the aggressive lossy profile in CI.

[krassowski]

I think that media-compression.lock adds too much overhead to maintaining this repository (adding new examples/modifying existing ones). Could we only run exclusively lossless compression instead? At least with optipng this should stabilise at a stable binary form.