Summary:
Commit 188491d3 introduced a regression
in ShellClient and broke the QuickTiling test - sorry about that.

This change fixes the failing test. The main reason was a slight change
in semantics of setGeometry. But the real problem is/was that the
geometry update blocking is not supported in ShellClient.

This change brings in the checks from Client to ShellClient, thus that
the geometry is properly blocked and when unblocked the geometry is
updated correctly or a request is sent. What makes it rather difficult is
that the geometry update blocker operates on an updated geom while
ShellClient::setGeometry needs to compare to the original geom to
determine whether a direct update or a request is required.

Thanks to supporting geometry update blocking the changeMaximize calls
are adjusted to use setGeometry instead of requestingGeometry directly.

Overall this is an important improvement and might also allow us to no
longer require the special RequestGeometryBlocker in ShellClient.

Test Plan: Tests pass

Reviewers: #plasma, #kwin

Subscribers: plasma-devel, kwin

Tags: #kwin

Differential Revision: https://phabricator.kde.org/D8956