Some follow-up test fixes after seeing effect in PB2

Rewrote logic (first commit was incomplete, sorry)

Apparently caused by logging to table
Turn on logging to file only, add to .opt file for tests needing log to table

Moved some dynstr_free() further up

Enabled proper pattern for Warnings and ERRORs
Added some suppressions

Moved some dynstr_free() further up

Quicker test collection and better output with --start[-dirty]

Fixed in two comments as well

Adding @<platform> syntax

Output would match an empty result file but we don't check
Allow empty output IFF there is an empty result file.

Set to one week for testcase and suite timeout
Also set one day timeout for PID file creation (not currently needed in 5.1 but might become, and is needed in azalea)

Variable name mismatch
Map variable have_innodb=YES to innodb=ON

Bug in Perl
Scrap attempt to do this smartly on AIX, just drop the test and assume it's OK
This commit undoes the previous push and adds a line to ignore on AIX

Bug is actually in Perl
Fixed by trapping and ignoring error from IO::Socket::UNIX

Small amendment to original fix, as it did not work in azalea
Need to handle combinations, would eventually break in 5.1 too

Also fixed mysqld.cc to avoid popup-boxes

Change of variable states in Bug 19027 was not backported to v1
Changed ne "TRUE" to eq "OFF"

General problem: some test cannot run in V1, expect more in future
Implement general mechanism for listing incompatible tests

Server args containing spaces do not work on Windows
Fixed my_safe_rprocess-win to re-apply "" around such args

Fix the name....

Updated to use general_log[_file] and slow_query_log[_file]

Added search for example plugin in lib/mysql/plugin

Let the user specify 'auto' for parallel value
Also set --parallel=auto in default.push so we can get this tested.

Added calls to mtr_report_stats() also after timeout or too many failures

Adding "w# " before the test result
Also enable --parallel=2 so this can be tested.