feat(validator): add SYNC_MODE so a fresh validator can warp sync

[nol4lej]

Closes #6 - same idea, opt-in instead of hardcoded.

--sync warp lets a brand new validator prove finality and fetch the latest state instead of replaying every block. Two things keep it from being a safe default:

• On a volume that already holds finalized state the node ignores it: Can't use warp sync mode with a partially synced database. Reverting to full sync mode.
• When it does apply, the validator isn't an archive node, so the gap is backfilled headers only — no block bodies before the sync point, for good.

So it goes behind ${SYNC_MODE}, empty by default, same pattern as PUBLIC_ADDR and RESERVED_NODES. Bootstrapping a new validator sets SYNC_MODE=--sync warp; existing ones are untouched.

Validators only. rpc and indexer-rpc run --blocks-pruning archive, and an archive node exits at startup with Warp sync doesn't work for archive nodes and crash-loops under restart: unless-stopped.

Verified against ghcr.io/orbinum/node:testnet-latest (0.2.0): empty ${SYNC_MODE} leaves no stray args, --sync warp splits correctly, warp engages on a fresh volume, is ignored once finalized state exists, and is fatal on an archive node.