feat(experimental): add luDF reductions and histograms

[ibgreen-openai]

Goals

Add GPU-resident global dataframe reductions and filtered numeric histograms to luDF without implicit batch packing, command submission, or CPU row readback.

Changes

• Add typed frame.aggregate(definitions) / query.aggregate(definitions) global count, sum, min, max, and mean operations.
• Preserve float32, sint32, and uint32 source types for sum/min/max; expose float32 means and uint32 counts with one-row GPU-owned result tables.
• Add frame.histogram(column, {bins, domain}) and {edges} variants; keep domains explicit so filters and nullable rows cannot corrupt inferred bounds.
• Compose filters, chained derived columns, encoder-ordered mutable parameters, preserved source batch/chunk masks, and canonical source ownership with the existing graph extension.
• Reuse GPUReduction, GPUHistogram, and GPUMask; exclude null/NaN/infinite statistic inputs and expose explicit GPU validity flags for every nullable scalar result.
• Preserve integer wrap semantics, reject unsupported count/buffer limits, retain native signed/unsigned histogram domains, and safely synthesize empty float NaN payloads at runtime for Dawn/WebGPU compatibility.
• Add strict immutable Node API/type/validation contracts and real-WebGPU CPU-equivalence cases for all scalar formats, zero batches, filtered/derived inputs, NaN/infinity, irregular edges, repeated encoding, and destruction ordering.

Verification

• nvm use — Node 22.22.1.
• yarn install — the existing enterprise registry returns HTTP 403 for an upstream alpha dev-tool package; reused the previously verified ignored workspace dependency tree.
• yarn lint fix — passed after final source changes.
• yarn build — passed across every workspace package after final formatting.
• Focused luDF Node suites — 54 tests passed.
• Focused real Chromium/WebGPU — six integration scenarios passed.
• CI=1 yarn test — Node: 597 passed and one skipped; real Chromium/WebGPU: 1,576 passed and 25 skipped.
• yarn website:build — passed, including validation of 461 documentation pages.
• (cd website && yarn build) — passed, including validation of 461 documentation pages.
• yarn examples:typecheck — passed across all 46 example workspaces.
• yarn bundle-size — passed all seven existing bundle budgets.

Risks and limitations

• Existing GPU reduction and histogram primitives operate on native 32-bit scalars; floating-point sums and means have float32 precision, and signed/unsigned sums preserve native 32-bit overflow semantics.
• Domains and irregular edges must be explicit; automatic filtered extents are deliberately unsupported.
• Scalar result tables contain one merged row, histogram tables contain one dense result batch, and original row masks/indices/counts retain source batch topology.
• Stable sorting, top-K, limited joins, and interactive documentation/examples follow independently.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 030dece020

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".