fix(linux-clickhouse): read the process list before the query log

[datlechin]

Why

A ClickHouse write that ran past its deadline could resolve as Unresolvable even though it applied. That is the Docker test flake from #3437. The resolver read system.query_log, then system.processes. An ALTER that finished between the two reads was in neither.

Change

• Read system.processes first, then flush and read the log. The server queues a statement's last log entry before it takes the statement off the list, so a statement that ends between the reads is either still listed or has its end in the flushed log.
• A login without a grant on system.processes resolves from the log alone instead of failing with ACCESS_DENIED. ClickHouse 25.8's stock config sets select_from_system_db_requires_grant.

Verified

• New an_alter_that_ends_between_the_list_and_the_log_is_applied holds the ALTER with SYSTEM STOP MERGES and lets it finish between the two reads. The old order answered Unresolvable there.
• New resolve_refused_the_process_list_answers_from_the_log failed on the old code with error 497.
• The ClickHouse Docker suite ran 6 times. The MySQL 8.4 and MariaDB 11.4 DDL suites pass.

Not in this PR

• commit_with_no_record_is_unresolvable and calls_share_one_call_options each failed once while a build ran beside the suite, and passed in every later run.