Fix bulk card image downloader to use Scryfall CDN

[phughk]

Fetch card images from the Scryfall CDN instead of the rate-limited API

Summary

Card image downloads (desktop, mobile, and the bulk image browser) currently go through
api.scryfall.com, which rate-limits and is the main source of failed/slow downloads. Scryfall
also serves the same files unthrottled from cards.scryfall.io, but that requires already
knowing the card's UUID.

This PR adds a local, disk-backed cache (CdnUuidCache) that maps (set, collector number, lang)
→ Scryfall UUID, populated on demand from Scryfall's search API (ScryfallSetSync), and prefers
CDN URLs built from that cache over the old API path. The API and the existing cardforge-hosted
server remain as fallbacks, so nothing regresses if a card isn't in the cache yet.

How image resolution works now

Request timeline for one card face (cache miss on first call, hit on a later call once the
background sync has landed):

sequenceDiagram
    participant Fetcher as ImageFetcher
    participant Cache as CdnUuidCache
    participant Disk as cdn_uuid/*.json.gz
    participant Sync as ScryfallSetSync (bg thread)
    participant API as api.scryfall.com

    Fetcher->>Cache: getCdnUrl(set, cn, lang, face)
    Cache->>Disk: read set file
    Disk-->>Cache: no entry for cn/lang
    Cache->>Disk: record timestamped miss
    Cache->>Sync: queue set for sync (async)
    Cache-->>Fetcher: null
    Fetcher->>Fetcher: fall back to Scryfall API / cardforge URL

    Sync->>API: GET /cards/search?q=set:...
    API-->>Sync: card pages (paginated)
    Sync->>Disk: merge cn/lang → uuid

    Note over Fetcher,Disk: next request for the same face
    Fetcher->>Cache: getCdnUrl(set, cn, lang, face)
    Cache->>Disk: read set file
    Disk-->>Cache: uuid found
    Cache-->>Fetcher: cards.scryfall.io URL
    Fetcher->>Fetcher: download from CDN

Loading

Lookups never block on network: a miss is served from fallback immediately, and the sync runs on
a background thread pool. Unresolved (cn, lang) pairs are negative-cached with a timestamp so
they aren't retried on every lookup, only after a day.

How the pieces connect

flowchart LR
    subgraph Callers
        IF[ImageFetcher]
        GD[GuiDownloadFilteredCardImages]
        UI[CardImageBrowserScreen]
    end

    subgraph Cache Layer
        CUC[CdnUuidCache]
        SSS[ScryfallSetSync]
        Disk[(cdn_uuid/*.json.gz)]
    end

    subgraph Downloaders
        SW[SwingImageFetcher]
        LG[LibGDXImageFetcher]
    end

    IF -->|getCdnUrl| CUC
    GD -->|getCdnUrl| CUC
    UI -->|clearCache| CUC
    CUC <--> Disk
    CUC -->|queueSync| SSS
    SSS -->|merge results| CUC
    SSS --> ScryfallAPI[api.scryfall.com/cards/search]

    IF --> SW
    IF --> LG
    SW --> CDN[cards.scryfall.io]
    SW --> ScryfallDL[api.scryfall.com/cards/*]
    SW --> Cardforge[cardforge hosted server]
    LG --> CDN
    LG --> ScryfallDL
    LG --> Cardforge

Loading

Changes

• CdnUuidCache (new) — thread-safe, gzip-JSON-backed cache; resolves a UUID, or queues a
  background sync and returns null.
• ScryfallSetSync (new) — pulls one set's cards from Scryfall's search API (paginated,
  rate-limit-respecting), reading each face's own image_uris so double-faced cards with distinct
  front/back art resolve correctly.
• ImageFetcher.addScryfallUrl / GuiDownloadFilteredCardImages.buildUrl — try CDN first,
  then Scryfall API, then cardforge.
• SwingImageFetcher / LibGDXImageFetcher — recognize CDN URLs alongside API URLs so
  existing border/rounding fixups still apply.
• CardImageBrowserScreen — adds a "Clear CDN Image Lookup Cache" button.
• New gson dependency in forge-gui/pom.xml for cache (de)serialization.
• New i18n strings across all bundled languages for the clear-cache button/dialogs.

Testing

• CdnUuidCacheTest — URL formula, language fallback, DFC face resolution, miss recording/retry,
  cache upgrade on real data.
• ScryfallSetSyncTest — single/double-faced resolution, pagination, 404 handling, never
  overwriting existing entries, miss retry staleness.

For reviewers — long-term maintenance concerns

• New runtime dependency on Scryfall's search API: ScryfallSetSync is a second, independent
  integration point with Scryfall beyond the existing per-card download API. If Scryfall changes
  its search response shape or CDN URL format, both ScryfallSetSync.addCard/uuidFromUrl and
  CdnUuidCache.cdnUrl's hand-built URL formula need to be checked — cdnUrl does not verify
  the URL it builds actually resolves.
• Cache correctness depends on mergeSetEntriesWithFaces never overwriting real entries: this
  is the invariant that keeps a bad/stale sync from clobbering a working UUID. Any future change
  to the merge logic should preserve "real data always wins over a miss, and existing real data is
  never replaced."
• On-disk cache format has no version field: {cn: {lang: uuid | [front, back] | {miss}}} is
  read directly with no migration path. A future format change needs either a version bump or to
  stay a strict superset of the current shape, otherwise old cache files silently parse wrong.
• Silent local cache growth: one gzip file per set is written under
  {cacheDir}/cdn_uuid/ and only removed via the manual "Clear CDN Image Lookup Cache" button.
  There's no automatic pruning or staleness eviction beyond the per-entry miss-retry window —
  worth confirming this is acceptable for long-running installs.
• Background sync threads use the shared ThreadUtil service pool: syncPendingSets is
  fire-and-forget with no cap on concurrent set syncs; a user browsing many unseen sets in quick
  succession could queue many simultaneous Scryfall requests. Rate limiting is per-request
  (REQUEST_INTERVAL_MS) within a single sync, not across syncs.

[github-actions]

This PR has not been updated in a while nad has been marked on stale. Stale PRs will be auto closed

[phughk]

Will tackle tonight or tomorrow night

[phughk]

For context

• the other prs are closed (forge-extras mirror, cli tool)
• local only cdn uuid resolution
• resolution comes from api instead of bulk exports
• cdn resolution is non-blocking
• pr message has been re-written to cover details

[churrufli]

I'll happily rewrite my code #11572 once your PR gets merged — at the end of the day, all I wanted was for users to be able to download cards in their own language.

I also think the unique cards behavior I mentioned is interesting: when "unique cards" is on and that same edition has the card printed in more than one language, I'd like the one downloaded to be the one in the user's language.

And having the download language as its own setting is interesting too — a user might have the UI in one language but prefer to download cards in a different one, so it makes sense to let them choose that separately.