[Performance][P0] Add responsive cover pipeline and resilient IPFS delivery

[knzeng-e]

Goal

Make artwork appear immediately and consistently without compromising IPFS-backed content addressing.

Problem

The upload API currently accepts multi-megabyte original covers and pins them unchanged. The UI reuses those assets across hero, rooms, stage rail, player, and catalog surfaces. Gateway fallback occurs only after an image error; a slow gateway may stall without failing.

Scope

• Transform cover uploads server-side with sharp or equivalent.
• Produce a tiny placeholder plus 64, 160, 320, and 640 px AVIF/WebP variants.
• Extend the release manifest/read model with responsive cover refs and intrinsic dimensions.
• Use srcset, sizes, decoding="async", fixed aspect ratios, and lazy loading outside the hero.
• Use fetchpriority="high" only for the active hero/player artwork.
• Add gateway timeout/racing and remember the winning gateway per CID/session.
• Put an immutable caching edge/read-through gateway in front of IPFS where allowed.
• Preserve the original cover as an archival asset, not the default UI asset.

Performance budgets

• Hero cover p75 < 1 s cold on mobile 4G.
• Catalog thumbnail transfer target < 40 KB each.
• Zero layout shift caused by artwork.
• Broken/slow gateway recovery < 1.2 s before trying the next source.

Acceptance criteria

• Newly uploaded releases have responsive image variants.
• Existing releases degrade to generated aura placeholders while legacy covers load.
• Repeated surfaces do not independently download full-size originals.
• Cache headers are immutable by CID.
• Browser tests cover slow, failed, and CORS-rejected gateways.