Closes #86
Closes #80

Co-Authored-By: Robert Frenken <robert.frenken@proton.me>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Builds on top of #552. Adds the component-flavored `<Hexbin>` (Approach
B from `HEXBIN_DESIGN_COMPARISON.md`, kept in my working tree) — a
Svelte component that bins 2D scatter data into a pixel-space hex
lattice and renders each bin as a regular hexagonal cell. Pairs with the
existing `<Hexgrid>` at matching `binWidth`: a default `<Hexbin />` +
default `<Hexgrid />` tile the same lattice without user coordination.

## Why a component, not just the transform

The transform-flavored `hexbin()` (already in #552) cannot bin in pixel
space because it runs at template-evaluation time, before scales exist.
Three structural symptoms followed:

1. Cells distort under non-1:1 axis aspect ratios (transform bins in
data units; projected cells aren't regular hexagons).
2. Lattice drift between the transform and `<Hexgrid>` under `nice`,
`zero`, or explicit `domain` (transform reads `extent(rawData)`, grid
reads `scale.domain()`).
3. User has to coordinate `bins` between the transform call, `<Hexgrid
bins>`, and `r` on the Dot mark to get a coherent picture.

`<Hexbin>` runs as a child of `<Plot>`, calls `usePlot()`, and builds
its lattice in pixel space after scales exist. All three symptoms
collapse: cells are regular by construction, the lattice is
single-sourced, and there's nothing to coordinate. The transform stays
for users who want the function-API ergonomics.

## What's in this PR

- **Shared lattice helper** at
`packages/svelteplot/src/helpers/hexLattice.ts` (102 lines, 7 helper
tests). Both the existing transform and the new component import from
here. `transforms/hexbin.ts` refactored 176→142 lines,
`marks/Hexgrid.svelte` 92→73 lines — behavior unchanged, just
deduplicated math.
- **`<Hexbin>` component** at
`packages/svelteplot/src/marks/Hexbin.svelte`.
- **9 unit tests** in
`packages/svelteplot/tests/hexbin.test.svelte{,.ts}` — including a
sub-pixel center-alignment test (#6) between Hexbin and Hexgrid that
catches a Y-origin bug fixed in this PR.
- **3 examples** under `src/routes/examples/hexbin/` with light + dark
VR baselines under `src/snapshots/hexbin/`.
- **Showcase docs** at `src/routes/marks/hexbin/+page.{md,ts}` + sidebar
entry.
- **PlotDefaults wiring** so users can configure default Hexbin props
via `setPlotDefaults`.
- **`NEW_MARK_CHECKLIST.md`** at the repo root — checklist for adding
new marks with monorepo paths (rebased from a doc on my main branch).
The commit message has each box ticked with file:line refs.
- **Pre-existing tooling fix:** `scripts/generate-api.js` had 13 stale
`src/lib/...` paths from before the monorepo refactor (`3a1d5205`) —
updated to `packages/svelteplot/src/...` so `pnpm docs:api:marks` works
again.

## Bug fixes inline

Two real bugs found and fixed during this work:

1. **Color scale was always "unknown".** `reduceOutputs` writes its
result to `__fill`/`__stroke` (with the `__` prefix, see
`helpers/reduce.ts:113`), not `fill`/`stroke`. The Symbol-copy step in
the component was reading the wrong key, so every bin rendered as
`#cccccc99`. Caught by the count-reducer test.

2. **Hexbin/Hexgrid alignment was off by `binWidth/√3`.**
`Hexgrid.svelte:53` and `transforms/hexbin.ts:85` use `originY = mt`;
the original Hexbin used `originY = mt + binWidth/√3` to "push cell
(0,0) inside the rect" — but that breaks symmetry with the X half-pitch
offset and breaks alignment with Hexgrid by exactly that amount. Caught
by the alignment test (#6) — without it, screenshots looked plausibly
close to aligned.

## Tests

- 870/870 tests pass (861 baseline + 9 new for Hexbin).
- `pnpm check`: 0 errors.
- VR baselines were generated on a SLURM compute node; on a login node
the Vite dev compile was too slow for puppeteer's default 60s navigation
timeout.

The full commit message has the complete `NEW_MARK_CHECKLIST` with
file:line citations for every box.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Two fixes for #552's `feat/hexbin-transform` branch:

**Prettier formatting** — three Hexbin example files exceed the
`printWidth: 60` override for `routes/examples/**/*.svelte`; ran `pnpm
format`. Fixes the lint failure.

**Per-facet binning** — `<Hexbin>` inside a faceted plot was sharing one
bin map across all facets, so each panel rendered the union of all data.
Records are now partitioned by `(fxVal, fyVal)` before binning. Mirrors
`Density.svelte:272-286, 441-497`. New test in
`tests/hexbin.test.svelte.ts`.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>