Add retry logic with exponential backoff for Stellar RPC errors

[AbelOsaretin]

What

sendTransaction and getTransaction calls in src/lib/stellar.ts have no retry-on-network-error logic. Transient failures (timeout, 503, network blip) cause immediate transaction failure.

Why

Stellar RPC endpoints can experience transient failures. Without retries, a single network hiccup causes a transaction to fail even though it may have succeeded on-chain.

Scope

• Add retry wrapper for Stellar RPC calls
• Exponential backoff with jitter
• Configurable max retries and base delay
• Only retry on transient errors (timeout, 503, network error)
• Do NOT retry on permanent errors (400, 422)

Acceptance Criteria

• RPC calls retry on transient errors (up to RPC_MAX_RETRIES)
• Exponential backoff with jitter between retries
• Base delay configurable via RPC_RETRY_BASE_MS
• Permanent errors are not retried
• Logs each retry attempt

Technical Context

• src/lib/stellar.ts:34 — sendTransaction
• src/lib/stellar.ts:43 — getTransaction
• src/lib/registry.ts:53 — simulateTransaction
• New env vars: RPC_MAX_RETRIES (default: 3), RPC_RETRY_BASE_MS (default: 1000)