M1.3 OPFS headers store + resume + tip reorg

[melvincarvalho]

Part of #1 (M1).

Goal

Persist validated headers to OPFS; resume from the stored tip on reload; handle small reorgs at the tip.

Benchmark / acceptance

• Kill and resume mid-sync with no re-validation of already-stored headers.
• Report storage footprint for the full header chain.
• A tip reorg is detected and resolved to the most-work chain.

[melvincarvalho]

Done (headless, on testnet4).

• src/store/header-store.js — HeaderStore interface + MemoryHeaderStore + FileHeaderStore. Shared index (hash→height, cumulative work), exponential block locator, truncate for rollback.
• src/chain/header-sync.js — HeaderSync: resume from store, validate each header incrementally, reorg by the most-work rule, skip re-served prefixes. Depends only on the store interface + engine + a transport-agnostic fetchHeaders(locator).

Tested: 10 headless unit tests (store index/truncate/locator/file round-trip; sync fresh/resume/prefix-overlap/heavier-fork-adopted/lighter-fork-rejected), all green. Live testnet4: fresh sync 140k headers all validated; resume from disk in 0.3s; tip cross-checked over p2p.

Browser port is now two swaps: FileHeaderStore → OpfsHeaderStore, TCP Peer → WS-bridge Peer.