router.callro doesn't fallback to master when replica is in OperationError (regression after upgrade)

[vakhov]

Description:
After upgrading vshard to 0.1.35, Cartridge integration tests started failing in a scenario where the replica is in OperationError.
Previously, router.callro in this case returned data from the master, but now it returns box.NULL.

Steps to reproduce:

1. Start a cluster with master and replica.
2. Write data using router.callrw.
3. Put the replica into OperationError state (simulate failure).
4. Attempt to read the data via router.callro.

Expected result:
router.callro falls back to master and returns the data.

Actual result:
router.callro returns box.NULL.

Notes:

• Switching to router.callre fixes the issue.

• This behavior change was unexpected and breaks Cartridge tests:
  https://github.com/tarantool/cartridge/actions/runs/16872655825/job/47794894363?pr=2348#step:17:1989
  https://github.com/tarantool/cartridge/blob/871bb4d5a43d2ab2964475f92ab346f773abbb59/test/integration/vshard_storage_disabling_test.lua#L189

• Reported in Cartridge team and confirmed with vshard developers that this is not intended behavior.