enhancement: missing retry logic for EVM RPC and callback/webhook delivery

[user1303836]

Summary

Two areas lack retry logic for transient network failures:

1. EVM `fetch_logs()` (`adapters/src/evm.rs:91-117`)

Single RPC timeout fails the entire backfill. No exponential backoff or retry budget.

2. Callback delivery (`api/src/main.rs` `fire_callback()`)

Fire-and-forget with a 10-second timeout. If the webhook endpoint is temporarily unavailable, the callback is permanently lost with only a warning log.

Impact

• EVM: A single transient RPC failure aborts a multi-hour backfill that must be restarted from scratch
• Callbacks: Users have no visibility into whether their job completion callback was delivered; they may miss important notifications

Recommended Fix

1. Add exponential backoff retry (3 attempts, 100ms/200ms/400ms) to `fetch_logs()`
2. Add retry logic with backoff (3 attempts, 500ms/1s/2s) to `fire_callback()`
3. Don't retry on 4xx responses (client error), only on network/5xx errors

🤖 Generated with Claude Code