perf(home): establish field Core Web Vitals baseline for / and compare with /dashboard

## Problem

The marketing homepage `/` is now a pre-rendered, media-rich acquisition surface with a different runtime and user journey from the WebGL dashboard. We have mature field-CWV ownership for `/dashboard` in #4487 and its metric-specific children, but no equivalent route-level baseline for `/`.

External audits can therefore make unmeasured claims about homepage SEO/performance, and origin-level aggregates may hide route differences.

This issue is measurement-first. It must not assume `/` is failing, and it must not duplicate `/dashboard` remediation already owned by #4487.

## Scope

- Establish route-level field CWV for `/` on mobile and desktop.
- Compare against current `/dashboard` values from #4487 and its children using the same time window and metric definitions.
- If field coverage for `/` is insufficient, document that explicitly and use existing RUM (or add the minimum route attribution needed) rather than substituting a one-off Lighthouse score.
- Use lab runs only for diagnosis and reproducibility, not as the field pass/fail verdict.
- No performance fixes in this issue; file evidence-backed follow-ups only after the baseline exists.

## Measurement contract

For both `/` and `/dashboard`, segmented by phone and desktop where available:

- CrUX URL-level 28-day p75 and percentage-good for LCP, INP, and CLS.
- DebugBear RUM p75 and sample count by route for a matching recent window, if instrumented.
- One reproducible mobile lab profile recording the LCP element, LCP/FCP/CLS/TBT, transfer bytes, and request count.
- State data source, date window, sample count, and whether each value is field or lab.

## Acceptance criteria

- [ ] A comparison table for `/` vs `/dashboard` is posted with dates, device segment, p75, percentage-good, and sample availability.
- [ ] Mobile is treated as the primary decision segment; desktop is retained as a regression check.
- [ ] Origin-level and URL-level data are not mixed.
- [ ] The homepage LCP element and largest transfer or long-task contributors are identified in the reproducible lab profile.
- [ ] The result states one of: healthy/no action; insufficient field data/continue measuring; or a ranked evidence-backed follow-up.
- [ ] Any `/dashboard` findings are routed to #4487 or its existing child issues rather than duplicated here.

Start with the existing field-CWV work in #4487 and its metric-specific children, then gather matching CrUX and DebugBear RUM data for / and /dashboard. Record mobile and desktop segments, dates, samples, p75 and percentage-good values, and use one reproducible mobile lab profile for diagnosis. Done means a route comparison table, identified homepage contributors, and a conclusion that routes any dashboard findings back to #4487.

Written by the indexing model from the issue text.