sql: add separate log channel for statements that fail with errors

[michae2]

It's occasionally useful to examine all statement executions that failed with an error. (Similar to examination of a dead letter queue to diagnose problems.) We have a couple ways to do this today, but all are painful:

1. Log all statement executions to SQL_EXEC by setting sql.trace.log_statement_execute or sql.log.all_statements.enabled. This is a firehose and not suitable for production clusters.
2. Log slow executions to SQL_PERF (a.k.a. the slow query log) by setting sql.log.slow_query.latency_threshold and hope that some of the slow executions are also failures. This misses any failures under the threshold.
3. Turn on slow query tracing with sql.trace.stmt.enable_threshold or sql.trace.txn.enable_threshold. This is also not suitable for production clusters as it imposes a performance penalty on every query execution, and it might not necessarily find statement failures.

It would be much more straightforward to create a log channel devoted to statement execution failures.

Jira issue: CRDB-48089