feat(portfolio): surface degraded/stale chains instead of silent $0

[BitHighlander]

Problem

The vault discarded Pioneer's portfolio meta, so a chain that failed its fresh fetch (node down/timeout) or served cache older than 5 min showed $0 or a stale value with no way to distinguish "data unavailable" from "genuinely empty." No retry, no staleness signal.

Pioneer already emits the needed signal — verified live on api.keepkey.info (server release/v1.3.115): POST /portfolio returns meta.{degraded, degradedCount, failures[], staleChains[]}. The vault just threw it away in unwrapPortfolioEntries.

What changed

Backend (src/bun/index.ts)

• unwrapPortfolioResponse keeps meta per chunk; mergeMetas combines it across chunks.
• After each getBalances, push degraded + stale chain display-names to the webview via the existing pioneer-error channel — severity: 'warning' for soft faults, 'none' to clear. Degraded = server meta.failures ∪ chains whose pubkeys all landed in a failed chunk.

Schema (src/shared/rpc-schema.ts)

• Extend pioneer-error with optional severity + degradedChains/staleChains/staleMinutes. Missing severity stays a hard error — back-compatible with existing senders.

Frontend (Dashboard.tsx)

• Amber soft-fault banner with its own state, separate from the rose hard-error banner (hard error supersedes). "Sync now" + "Dismiss".
• Window-focus refetch when idle > 5 min.
• Degraded-chain background retry with 30/60/120/300s backoff; resets on a clean fetch.

i18n — English strings only; other 14 languages fall back via fallbackLng: 'en'.

Scope notes

• Reuses the existing banner channel rather than adding a parallel portfolio-fault event (one source of truth for "balance data problem").
• The retry triggers live in the webview, not the Bun process (the handoff placed window listeners in src/bun — Bun has no window).
• Dropped the handoff's onlyCaips targeted refetch (full refresh is ~5-6s — not worth the added branching) and the rest-pioneer.ts change (that's the external SDK surface, unrelated). Per-row stale indicators deferred.

Test plan

• tsc --noEmit clean (verified locally — 0 source errors)
• Degraded banner names the chain when a node is down (e.g. Optimism) instead of silent $0
• Stale banner shows "last synced N minutes ago" on >5min cache
• "Sync now" forces a refresh and clears the banner on success
• Idle >5min, refocus window → background refresh fires
• Degraded chains auto-retry on backoff ([Dashboard] Retrying degraded chains in console)