effects: taint `:nothrow` effect on unknown `:static_parameter` (conservatively)

[aviatesk]

With this commit, we taint :nothrow effect property correctly on access to unknown :static_parameter, e.g.:

unknown_sparam_throw(::Union{Nothing, Type{T}}) where T = (T; nothing)
@test Core.Compiler.is_nothrow(Base.infer_effects(unknown_sparam_throw, ((Type{Int},))))
@test !Core.Compiler.is_nothrow(Base.infer_effects(unknown_sparam_throw, ((Nothing,))))

This commit implements a very conservative analysis, and thus there is a room for improvement still, e.g.:

unknown_sparam_nothrow(x::Ref{T}) where {T} = (T; nothing)
@test_broken Core.Compiler.is_nothrow(Base.infer_effects(unknown_sparam_nothrow, (Ref,)))

fix #46771

[aviatesk]

@Keno is it okay for now to merge this as is, and try to improve the accuracy of this nothrow analysis in a follow up PR (it seems a bit tricky than I expected though)?

[Keno]

I would prefer to try to get at least some precision in at the same time, because I suspect we're relying on this bug for performance in a number of applications

[staticfloat]

@aviatesk I rebased this branch, so that we can push forward Keno's PR more easily.

[vchuravy]

Will this need to be backported to 1.9 since it's a bugfix to the effect system?

[aviatesk]

Will this need to be backported to 1.9 since it's a bugfix to the effect system?

Yes, I think we should do it.

[aviatesk]

@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

Your benchmark job has completed - no performance regressions were detected. A full report can be found here.

[oscardssmith]

given that there aren't any regressions here, good to merge?

[aviatesk]

There is one test failure on this PR, so don't merge now.

[aviatesk]

@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.