This is a problem similar to what was recently fixed in

- Bug#26926666 SPJ: 'LOOSESCAN' TABLES SHOULD NOT BE CHILD MEMBERS IN A PUSHED JOIN

Furthermore, there also existed a similar 'push limitation'  for
child tables using the 'FirstMatch' semi join strategy, probably since
the first release of the join pushdown feature.

After studying all these 3 limitations, we realize that the common denominator
for them all, is that a scan-child being part of a semi join 'nest', should
not be pushdown joined with a parent table not being within the same
semi joined 'nest'.

This patch takes advantage if this realization and thus implement
a single mechanism covering this pushability limitation for
both the existing 'FirstMatch' and 'LooseScan' strategy, as well
as the 'Duplicate weedout' being the subject of this bug report.

The abstract_query_plan method ::get_join_type() is extended
to also identify JT_SEMI_JOIN between a tableand its predecessor.

::is_pushable_as_child() replace its current logic to catch
'FirstMatch' and 'LooseScan' handling with checking for JT_SEMI_JOIN
instead, which now will also automagically catch 'Duplicate weedout'.

As a side effect a MTR testcase which previously incorrectly
rejected 'FirstMatch' pushing of two tables inside the *same*
semi join nest is now allowed to be pushed.

The 'explain' -> show warnings output is slightly changed as a
result of not specifically checking for 'FirstMatch' and
'LooseScan' anymore. Instead 'semi join' is reported
as the reject reason.