fix: stop a width the layout cannot draw, and say when two rows share an id

[ndlabdev]

Summary

Two defects reached the same place from different directions, so they travel together.

A column width that the layout cannot draw destroyed the grid, silently. A NaN or Infinity width reached the CSS custom property as NaNpx. A custom property accepts that, but grid-template-columns: var(...) then resolves to an invalid value, the declaration is dropped at computed-value time, every column folds into a single track and the cells stack down the page. Nothing threw and nothing was logged, and widthOverrides still read as a plausible record while the layout was already gone.

setState separately threw on a corrupt columns slice: an order that was not an array reached .filter inside the caller's own call.

Two rows sharing an id failed silently in the data: the row index keeps the last row for a repeated id, so an edit addressed to the row the user opened was written to a different one.

Closes #43
Closes #44

Changes

Where a number becomes CSS. buildColumnCssVars and columnTrackSize refuse a non-finite value and fall back to the column's minimum. This is the gate that closes the failure rather than a list of routes, and while writing it three more routes turned up that the issue had not named: a container measured as NaN, which makes every flex column NaNpx at once, and a definition written with width: NaN or flex: NaN. None of them pass through a setter or a snapshot.

Where an unusable value enters the model. setWidth and setWidths refuse a width that is not finite, where clamp and Math.round had been carrying NaN straight through. setWidth returns the width the column still has. These stay alongside the gate above: they keep junk out of the model rather than papering over it at the end.

Where a snapshot is read. resolveColumnSnapshot now reads values as carefully as it already read keys. Only string ids order the columns, only finite numbers set a width, only real booleans hide a column or fold a group, and a columns slice that is not an object is read as nothing. The fields are typed as unknown while they are read, because naming the type there would be the same promise that let the broken snapshot in.

Where rows are indexed. nodesById compares index.size against nodes.length, and a development build names the ids that collided. Production pays that one integer comparison for the whole data set; working out which ids repeated costs a second pass and only a build that will print it pays for that.

Behaviour changes

• A width that cannot be drawn no longer reaches the layout. A column given one keeps its minimum instead, so a column can come out narrower than intended where it used to take the whole grid down with it.
• setWidth returns the current width instead of NaN when handed one. It still returns 0 for a column that is not there.
• A corrupt snapshot value is dropped rather than applied, so a column falls back to what its definition says.
• One console.warn in development, the library's only console statement. no-console is disabled on exactly that line: an error would take an app down over data the grid can still draw, and there is no logger to route it to.
• No valid input changes. Healthy widths, flex weights, minimums and pin offsets are written exactly as before.

Checklist

• pnpm check: 1503 files, 0 errors, 0 warnings
• pnpm lint: clean
• pnpm test: 106 files, 1323 passed, 13 skipped
• pnpm bench: ran clean, and budgets.test.ts passes inside the suite. The added work is one comparison per data set and one finite check per track, not per row per frame
• The browser regression test asserts the computed grid-template-columns, not the override record. Backing the fix out turns it red with one track where there should be two, which is how the collapse was confirmed in the first place

Note

This branch and #42 both add to ## [Unreleased] in CHANGELOG.md, so whichever merges second needs a short conflict resolved there.