Port upstream 0.55.0: parallelize spend baselines

[Finesssee]

Summary

• port upstream CodexBar v0.55.0 spend-dashboard baseline parallelism from perf(spend): parallelize loads and memoize model build steipete/CodexBar#3105
• run independent Codex and Claude baseline scans concurrently while keeping each provider's own 7d/30d scans serial
• preserve the existing per-provider incremental cache assumptions and downstream spend contract semantics

Upstream evidence

• target release: v0.55.0
• upstream commit: 1cf98b330 / perf(spend): parallelize loads and memoize model build steipete/CodexBar#3105

Validation

• cargo test --manifest-path apps/desktop-tauri/src-tauri/Cargo.toml usage_spend --no-fail-fast -> compile/test pass
• focused rustfmt and git diff --check -> pass

Scope

Usage & Spend baseline load performance only. No provider-auth, UI, or version bump changes.

[Finesssee]

Thermo-nuclear review

No material code-quality findings. The concurrency boundary is placed at the independent provider level while each provider keeps its own 7d/30d scans serial, so the change removes unnecessary sequential orchestration without spreading shared mutable state.

Verdict: APPROVED.