[roll] Roll fuchsia [cqhci] Block task submission immediately for recovery

When a task fails and we need to run error recovery, immediately stop
accepting new tasks.  We used to wait until the async recovery task was
scheduled and it blocked the queue.

This issue was discovered while attempting to reproduce the linked test
flake in test_error_recovery, and I believe that this was in fact the
root cause of that flakiness: there was a small window where the reader
client could submit another request after needs_recovery was set but
before the recovery task blocked the queue, and that additional task
would also fail when recovery starts, leading to a spurious extra failed
task.

I modified the test harness to be explicit that we do not expect to
receive any requests after an error has been acked by the driver, but
before it has run recovery.  This least localize the failure if this
sort of issue happens again.

Original-Original-Fixed: 519586366
Multiply: cqhci-test
Original-Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/1645577
Original-Original-Revision: 7674d440d26d86505be64f098baf5e45ddbd3516
GitOrigin-RevId: f16e938584855abac71ba00985fbd7ce6c7356d0
Change-Id: I66d893ac0eb612bda27c7062506818fd4a1af503
Reviewed-on: https://fuchsia-review.googlesource.com/c/integration/+/1644176
Cr-Commit-Position: refs/heads/main@{#194482}