feat(api): add migrate_adapter rebalance keeper

[collinsezedike]

Summary

• Removes MeridianRouter entirely: per-user delegated rebalancing is structurally impossible on Stellar, token transfers and burns always require the token holder's own signature and there is no allowance/delegation primitive
• Adds a scheduled migration keeper (packages/stellar-sdk-helpers/src/migration-keeper.ts) that compares live rates across a vault's candidate protocol adapters and calls the existing, already-admin-gated migrate_adapter when a candidate clears a configured minimum improvement, with the same retry, deadline-budget, and structured-failure-reporting shape as the existing accrue keeper
• This is not yet functional against the live testnet vault, two independent gaps remain, both tracked separately rather than rushed: rate comparison is deliberately pluggable and defaults to always-unknown ("never migrate"), since neither Blend nor DeFindex's adapter exposes a ready-made comparable rate today ([Feature] Implement real rate sources for the migration keeper #511); and the live testnet vault predates migrate_adapter being added to vault/src/lib.rs and doesn't have the function at all, confirmed directly against the live contract ([Bug] Live testnet vault predates migrate_adapter, needs redeployment #514). See apps/docs/operations/migration-keeper.md
• Deploys and initializes a real MeridianDefindexAdapter on testnet (CAJVTA7EC3ZL3G4WSU4QIRB7RU7SUFUUJDEB7JE6CQQNPE7QC5OBSAM6) against the existing DeFindex testnet vault, so there's a genuine second candidate to evaluate once the gaps above close; it is not the vault's active adapter. Deliberately not wired into packages/shared/src/constants.ts, that file gates a required CI check verifying the vault address's on-chain bytecode against source, and [Bug] Live testnet vault predates migrate_adapter, needs redeployment #514 already fails it independent of this address; set MERIDIAN_DEFINDEX_ADAPTER_ID directly instead
• New api/v1/keepers/rebalance endpoint. MERIDIAN_MIGRATION_KEEPER_SECRET_KEY is a separate, higher-trust key from the accrue keeper's since migrate_adapter is admin-gated and accrue() is not. Its vercel.json cron entry is known to be non-functional on the Hobby plan (same limitation the existing accrue keeper's cron already hits), tracked separately in [Chore] Move keeper scheduling off Vercel Cron #513
• Extracts the accrue keeper's retry and transaction-submission logic into shared keeper-retry.ts/keeper-tx.ts modules so the new migration keeper reuses them instead of duplicating a third copy; closes out several items from [Chore] Deduplicate retry/error-handling logic in the accrual keeper #510 as a result (tracked there)

Test plan

• cargo build --workspace passes with the router package removed
• pnpm typecheck passes across stellar-sdk-helpers, api, shared
• pnpm test passes: 166 tests in stellar-sdk-helpers (18 new for the migration keeper), 35 in api (3 new for the rebalance endpoint)
• Manually verified the deployed MeridianDefindexAdapter on testnet via stellar contract invoke: get_protocol returns defindex, get_pool returns the configured DeFindex vault, total_assets returns 0
• Manually verified the live vault's actual callable methods via stellar contract invoke -- --help: no migrate_adapter, confirming [Bug] Live testnet vault predates migrate_adapter, needs redeployment #514

Closes #469

[vercel]

Deployment failed for project meridian with the following error:

Hobby accounts are limited to daily cron jobs. This cron expression (*/15 * * * *) would run more than once per day. Upgrade to the Pro plan to unlock all Cron Jobs features on Vercel.

Learn More: https://vercel.link/3Fpeeb1