sql: nil-out pointers in deallocated prepared statement

We saw a case recently where tracing enabled on a Prisma app caused Prisma to create a new prepared statement for every issued query. As far as I can tell, this is not due to an actual memory leak. Instead, the rate of allocations simply outpaced the go GC, eventually leading to 100% CPU and OOM during high QPS periods. In the past, we've handled situations like this by nil'ing out pointers in pointer-heavy data structures before losing the reference to the structure. We should do that for prepared statement memory.

Jira issue: CRDB-63297