[MNT] fix: PR CI trigger condition

[fkiraly]

Fixes an unreported bug - it would seem that CI is not triggered on PR by GHA.

This minimal changes that CI works as intended, on PR.

[fkiraly]

... why do we lint commit messages at a stage where they have already been committed?

I think a better solution is to allow only sqash commits (default PR title and preamble)

[fkiraly]

... & you don't like windows users? 😄

[fkiraly]

... the tests take kind of long, given that there are only a few estimators in here ...

I would recommend profiling and see how long which tests take.

[fkiraly]

FYI, I added a fail-fast: false flag to the matrix CI, so all elements run to the end even if one fails. Why: will run simultaneously anyway, so expected to terminate around the same time, shutting most of them down close to the end will waste the information on whether the failure is also on other python versions or OS.