Chunk GPU trace span storage

[ibgreen-openai]

What changed

• partition canonical trace spans into bounded GPU source chunks without repacking rows or splitting candidate batches
• run candidate visibility, density, picking, compaction publication, and indirect rendering across chunk-local span buffers while preserving stable global span IDs
• expose the 10M span option on portable WebGPU limits and automatically disable dependency rendering only when its endpoint lookup still requires a monolithic span binding
• report the active chunk count and chunk-aware capacity contract in the demo
• add CPU invariants and a forced multi-chunk WebGPU rendering test

Why

The 10M trace scale previously required one 320 MB span storage binding. Portable WebGPU adapters commonly cap a storage binding at 128 MiB, even though the rest of the trace pipeline is already GPU-driven. Bounded source chunks remove that single-binding ceiling without changing filtering, stable compaction, or indirect draw semantics.

Validation

• nvm use — Node 22.22.1
• focused Biome lint over all six changed files — passed
• focused trace viewer node suite — 14 tests passed, including WGSL parsing and chunk-boundary invariants
• yarn build — passed once after the implementation; the final repeat progressed through modules/experimental before the local session was interrupted
• yarn install — blocked by the local package firewall returning 403 for newly pinned packages
• yarn lint fix and yarn test wrappers — unavailable in the isolated worktree because that install could not complete; equivalent focused lint/tests were run directly

CI is the authoritative full lint/build/test gate for this PR.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2fc68925b5

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".