On some servers that:
- do have a heavy load in production
- do have very short sessions
- do have a very long uptime
the connection_id() counter can grow at a very high rate,
and eventually overflow the 32bits limit.

For the performance schema, a similar THREAD_ID counter
also exist, for table performance_schema.threads,
that is also implemented as a 32bits value.

This fix changes the following columns:
- performance_schema.threads.thread_id
- performance_schema.threads.processlist_id
to use BIGINT UNSIGNED (64 bits unsigned values)
All related columns in performance schema tables are changed accordingly.

The motivation for changing both thread_id and processlist_id at
the same time is to avoid un necessary schema changes later.

This fix changes the performance schema internal counter
for threads.thread_id to use a 64 bits value.

This fix does NOT change the internal counter for connection_id(),
which is done separately (see related bugs)

For clarity in the code, PFS_thread::m_thread_id has been renamed to
PFS_thread::m_processlist_id, to avoid confusion with m_thread_internal_id.