Dead lock may occur on MEMORY_STRICTLY mode when get connection on concurrency environment

[tuohai666]

In MEMORY_STRICTLY mode to avoid dead lock in connection pool, we must meet the condition:
ds.maximumPoolSize >= acceptor.size * ds.table_count.

If we get all connections needed synchronized between queries, we just need to meet the condition:
ds.maximumPoolSize >= ds.table_count.

According to the performance test, there's little degrade of the QPS.

[terrymanu]

The problem is SQL request may route to multiple sharding tables on same database, and each SQL request may get part of database connection, and waiting to release by others.

So just sync get connection as a atomic operate, let one SQL request can hold whole database connection at once.

fixed at 3.0.0.M3.