Summary

• Forward the server's lastFailure object (jobId, code, message, retryable, failedAt) in the output of cloudsync_network_send_changes() and cloudsync_network_sync() so callers can see why the
  server rejected an apply job.
• When cloudsync_payload_apply fails during the receive phase (e.g. unknown schema hash, invalid checksum), return structured JSON with a receive.error field instead of failing silently — applies
  to all three sync functions (cloudsync_network_send_changes, cloudsync_network_check_changes, cloudsync_network_sync).
• Endpoint/network errors (server unreachable, auth failure) always raise a SQL error. Processing errors (cloudsync_payload_apply failures) are returned as structured JSON. If you get JSON back,
  the server was reachable.
• Document the error-handling contract in API.md and as a code comment in network.c.

Test plan

• CI integration tests pass (Init+Sync, Enable Disable, Peer tests)
• Trigger a schema-hash mismatch and verify cloudsync_network_send_changes() includes send.lastFailure
• Trigger an apply error during cloudsync_network_sync() and verify the output includes receive.error with the error message and send fields are still populated
• Trigger an apply error during cloudsync_network_check_changes() and verify the output includes receive.error (not NULL, not a SQL error)
• Verify endpoint errors (e.g. bad URL) raise a SQL error in all three functions
• Verify normal sync cycles (no errors) produce the same JSON shape as before (no lastFailure or error fields)