remove obsolete filter from CI

[aplavin]

... also check whether tests fail on the upcoming julia version

[aplavin]

Indeed,

Accessors.jl/test/perf.jl

Line 181 in bd91513

@test b_right.time ≈ b_default.time rtol=0.8

fails (composition order).
It appears to be a genuine regression in Julia. But also not sure what's the relationship of this test to Accessors, why L vs R composition orders are compared in the tests here... @jw3126 do you remember?

[jw3126]

A long time ago in julia a composed lens was fast or slow depending on the composition order. E.g. these two had different performance:

@optic(_.d) ∘ (@optic(_.c) ∘ (@optic(_.b) ∘ @optic(_.a)))
((@optic(_.d) ∘ @optic(_.c)) ∘ @optic(_.b)) ∘ @optic(_.a)

The purpose of the test was to check that @optic _.d.c.b.a gives us the fast one. Then julia improved and both orders became fast. Now it seems there is a regression and we again have a fast and a slow order.
Probably we should change the test to be:
@test min(b_left.time, b_right.time) ≈ b_default.time rtol=0.8 # no matter which order is faster, it is the default
E.g. we only care to get the fast one, but we don't really care in the test which one is the fast one.

[aplavin]

Thanks for the background, updated the test accordingly and added a comment.