Adds a suffixStart option to parquetMetadataAsync that allows the caller to specify a byte offset to start fetching from (instead of fetching the last initialFetchSize bytes).

It's analogous to fetching bytes [idx:] (idx to EOF) instead of [-n:] (current behavior: last n bytes).

I built [a dashboard][awair] that reads from append-only Parquet files in S3:

Description of the PR...

Description of the PR...

Fixes a bug with the first pan/zoom after page load, when traces have different zorder values: some traces are not redrawn until the pan/zoom completes (while others are). All traces are redrawn continuously during subsequent pans/zooms, as expected. This is particularly noticeable in react-plotly.js, which calls Plotly.Plots.resize() after mount.

Root cause: When relayout is called (e.g., during resize), supplyDefaults resets _plots via linkSubplots, losing z-indexed subplots (e.g., xyz2, xyz3). Since relayout doesn't trigger drawFramework, they aren't recreated, causing updateSubplots in dragbox.js to miss them during pan/zoom.

Fix:

• In linkSubplots (plots.js): preserve z-indexed subplots from oldSubplots

Description of the PR...

• Add dffs-shell-integration command for convenient shell aliases with c (back-compat) and n (no-color) variants
• Add -C/--cached flag to git-diff-x for comparing HEAD vs staged changes
• Fix exit code propagation in git-diff-x and comm-x
• Improve test assertions: replace vague in checks with precise line-by-line comparisons
• Remove deprecated Tuple import, use built-in tuple
• Update CI workflow to match ghpr/git-didi pattern

Description of the PR...

Description of the PR...

Description of the PR...

Description of the PR...