[Gh-8589] A new approach to non-nullable typed associations for BC

[beberlei]

This partially reverts #8589 because it was overzealously working on non typed properties as well, defining 'nullable' => true on all join columns. This made CascadeRemoveOrderTest fail on databases with foreign keys.

With this approach the code about join columns stays untouched, unless the typed property is defined as non-nullable and join columns are used.

Fixes #8638

[beberlei]

@Lustmored ping for a comment :-)

[Lustmored]

It generally looks good and certainly solves the issue 👍 I was certain that nullable = true should be the default hence the regression.

I would just leave tests for targetEntity (per code comments) as they seem useful.

[rvanlaak]

I have a question while trying to understand the scope of this PR, and whether we will wait on 2.9 to release or try helping by contributing a PR.

Could this PR somehow be related to the long-standing #4090 issue from 2014 on OneToOne relationships having their FK set to non-nullable?

In other words, is 2.9 going to allow the scenario that PR #6533 tried to solve earlier? 👏 If not, would there be anything to help, and/or could a fix get backported to make it in 2.8.x ?