Makes the error-log available via a
performance_schema table ("error_log")
as a quality of life feature.

To include the log-events written during the
previous shutdown into the information the DBA
may review, the server will reload the previous
run's error log file at start-up.

This feature will allow for access to this
information to those who need it without
the need of creating an account for them
on operating system level and granting them
access via SSH / read privileges on file-
system level, or setting up monitoring tools
for them.

Additionally, this change will make writing
certain test cases simpler, as a .test may
now simply query a table rather than having
to parse a log file.

The new feature is largely implemented in
the following new files;

a) data collection (ring-buffer, sink adding log-events
to the ring-buffer in normal operation, and reading of
previous logs into the ring-buffer) happens in
log_sink_perfschema.cc.

b) Data presentation via the performance_schema happens in
table_error_log.cc and cursor_by_error_log.cc.

In an effort to keep the file layout logical and the logging
core of manageable size, the buffered and trad loggers have
been moved into separate files.

Approved by: Jens Even Blomsoy <jens.even.blomsoy@oracle.com>