by pushing the hypothesis controlled prng seed into scipy

But exclude 0-dim for now

This ensures no two spaces with different ids are used for same-dim
spaces

VectorArray generation is entirely routed through creation numpy
data first and creating a space from that.

The `implementations` decorator should be used for all vectorarray
tests. It abstracts the internal combination of
data + spaces in case we need to change it in the future.

also remove fasttest target, this hasn't been supported for a while now

These tests were all checking if `VectorArray.__getitem__` raises _any_
Exception given a "wrong" index. Nothing in these tests was actually
specific to `dot`, `almost_equal`, etc.

reduce example cout in other PR builds too

and rais the default deadline by 5x. Lower limit resultet in flaky tests
where deadline had passed in 1st call, but ont on repeats

more compositye strategies