fix: cache auto retry on MOVED / ASK / ...

[wyxloading]

try fix #700

DoCache, DoMultiCache end out with one or some redis transaction like:
MULTI, PTTL {key}, GET {key}, EXEC

When we scale a redis-cluster, PTTL {key} may return a MOVED / ASK error because slot have mark migrate / migrating / migrated to another node, then the transaction may fail.

We hope driver can handle this case automaticlly by just retry the transaction on node which error said.

EDIT:
We tested in our real test environment and it works.

[wyxloading]

BTW, happened to find out testcase TestClusterClient_SendToOnlyPrimaryNodeWhenPrimaryNodeSelected Do with no slot may fail sometimes. Seems like driver will pick any conn to do initSlot commands, but testcase assume it only use primary node conn

[rueian]

BTW, happened to find out testcase TestClusterClient_SendToOnlyPrimaryNodeWhenPrimaryNodeSelected Do with no slot may fail sometimes. Seems like driver will pick any conn to do initSlot commands, but testcase assume it only use primary node conn

Thank you for catching that. We can just delete all the Do with no slot test cases because we don't guarantee using primaries for those commands anymore. In the future, nodes may have mixed roles of primaries and replicas.

[wyxloading]

@rueian all review discussions resolved, please check this out.

[rueian]

Thank you @wyxloading!