connection timeouts and keepalive

[james-owen]

The main knex pool ran queries with no bounds: no statement or query timeout, keepAlive off, and no acquireConnectionTimeout (the createDBIfNotExists bootstrap client already had one), so a statement that never returns holds a pool connection with nothing to cancel it. Adds statement_timeout (300s, server-side) and query_timeout (330s, client-side) to the connection--the server cancels first so the error names the statement that overran--turns on keepAlive, and gives the pool the same acquireConnectionTimeout the bootstrap client uses. Overridable via CLAY_STORAGE_STATEMENT_TIMEOUT and CLAY_STORAGE_QUERY_TIMEOUT. Migrations are unaffected--postgres-migrations gets its own plain connection, not the knex pool. The search-detector's pg client has run against prod with the same 300s statement timeout all week.