Align OLAP transaction timeout with OLTP for tpcc-olap

The `tpcc-olap` workload fails ~43% of runs with sysbench aborting:
`FATAL: 'thread_run' function failed: (not a string)`. Plain `tpcc`,
which is byte-for-byte identical except for running vtgate in OLAP mode
(`--mysql_default_workload=OLAP`), passes 100% of the time.

Root cause: vttablet enforces a *separate* transaction timeout for OLAP
sessions via `--queryserver-config-olap-transaction-timeout`, which
defaults to 30s. We tuned up the OLTP timeouts long ago
(`transaction-timeout=300s`, `query-timeout=900s`) but never raised the
OLAP one. Under TPCC's lock contention (42 threads, 2 warehouses), an
OLAP-session transaction routinely blocks past 30s, so vttablet kills
the connection. A killed connection — unlike a retryable lock-wait error
that sysbench ignores via `--mysql-ignore-errors=all` — severs a
sysbench worker mid-run and crashes the benchmark.

Set `--queryserver-config-olap-transaction-timeout` to 300s so OLAP
sessions get the same headroom as OLTP, matching the regular `tpcc`
workload.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: arthurschreiber

ansible/roles/vttablet/templates/vttablet.conf.j2

@@ -37,10 +37,12 @@ EXTRA_VTTABLET_FLAGS="--alsologtostderr \
       --queryserver-config-query-timeout=900 \
       --queryserver-config-schema-reload-time=60 \
       --queryserver-config-transaction-timeout=300 \
+      --queryserver-config-olap-transaction-timeout=300 \
     {% else %}
       --queryserver-config-query-timeout=900s \
       --queryserver-config-schema-reload-time=60s \
       --queryserver-config-transaction-timeout=300s \
+      --queryserver-config-olap-transaction-timeout=300s \
     {% endif %}
     --grpc_max_message_size=67108864 \
     --db_charset=utf8 \