feat(rate): enforce cross-tab performance budgets

[d-sorgcodexagent]

Summary

• add a packaged, strict performance-budget manifest covering all nine React and nine PyQt visualization tabs
• enforce cold-open, resize-settle, stable-frame, post-settle movement, and browser CLS diagnostics in protected artifact probes
• move CPU-bound PyQt Plots computation behind a generation- and owner-bound killable subprocess with exact request/result authority and retained prior panes

Validation

• full web: 128 files / 1,070 tests
• focused Python authority: 70/70
• PyQt Plots async/GUI/render: 32/32
• PyQt all-tab DPI 1.0/1.5 performance/visibility: 1/1
• React three-viewport performance E2E: 1/1
• governance/assertion policy: 33/33; module-size and changed-test assertion gates green
• Python 3.12 / MyPy 1.13, Ruff/format, TypeScript, ESLint, production build, diff-check, and wheel resource inspection green

Scope limits

These are protected initial-production-state diagnostics, not arbitrary user-hardware qualification, result-workload qualification, approved visual goldens, persisted layout behavior, or formal assistive-technology evidence. Those remain open under #4433.

[dieterolson]

Blocker: this branch commits six orphan gitlinks

git ls-tree -r on this branch shows six entries at mode=160000 type=commit — git submodule references — under .codex-worktrees/:

.codex-worktrees/friction-factors-3659
.codex-worktrees/pr-3602-fix
.codex-worktrees/pr-3752-movement
.codex-worktrees/pr-3766-modern-robotics-dbc
.codex-worktrees/pr-3780-pressure-flow
.codex-worktrees/pr-3784-deterministic-te

There is no .gitmodules on the branch, so nothing defines where these submodules come from. They are scratch Codex worktrees that got committed as gitlinks. Landing them would break git clone --recurse-submodules and any checkout with submodules enabled, and they point at commits that may exist only in a local worktree.

This is corroborated independently: #4473's description states it is a "clean scoped rebuild" that excludes, among other things, "six scratch-worktree gitlinks" — the same six.

Scope observation

Measured against the union of main and the four campaign branches (#4446, #4447, #4466, #4473), this branch's tree of ~5,950 files contains no source paths that exist nowhere else — the only unique paths are the gitlinks above. So the real work here is modifications to files the campaign branches already carry, not new modules, and this reads as one layer of a stack rather than an independent change.

Path-level containment is not content-level containment, so this is not a claim that the work is already covered — the per-file content still needs checking before anything here is closed. Recording it now so the gitlinks are dealt with first either way.

Suggested next step

Remove the .codex-worktrees/ gitlinks from the branch (git rm --cached each path) before this leaves draft. Per the decision on the #4447/#4473 collision, the campaign is landing via #4446/#4447/#4466, so it is also worth confirming what this branch still adds on top of those once they merge.

No action taken on this branch.