configure feature calculation is needlessly complicated

The configure script translates

• -feature-foo to cmake -DINPUT_foo=ON
• -no-feature-foo to cmake -DINPUT_foo=OFF

This contradicts what one would expect and is contrary to what our configure-cmake-mapping.md document says.

Also, there's a pecularity in how -feature-system-foo switches are handled. Those are translated to cmake -DINPUT_foo=system, which is surprising at least.

We have code that calculates the value of feature foo from INPUT_foo if Qt is configured via the configure script, and we're even saving whether a feature value was evaluated using INPUT_foo to support toggling features with and without configure.

All this special handling could be removed if configure translated arguments like this:

• -feature-foo to cmake -DFEATURE_foo=ON
• -no-feature-foo to cmake -DFEATURE_foo=OFF

I don't see at the moment why we're doing this INPUT_foo dance at all.
The only reason I can come up with is: QtProcessConfigureArguments.cmake is a direct port of the qmake configure code, and that's how it was done there and then.