Correct exponent_max to conform to IEEE definition

[musm]

rebased from #24978

cc @JeffreySarnoff

[mschauer]

Looks good, also https://juliahub.com reports no additional use.

[musm]

will merge within 48 hours. Original PR was approved by @simonbyrne

[goualard-f]

I do not get it. In Julia 1.6.0, Base.exponent_max still returns the "wrong value", e.g. :

julia> Base.exponent_max(Float64)
1024

[simeonschaub]

This was merged after 1.6 was branched off. But perhaps this change should be backported?

[goualard-f]

I needed the correct definition but suffices to know for now that it will eventually make its way to the next release. In the meantime, I can rely on some workaround. Thanks.

[KristofferC]

This caused https://github.com/FedericoStra/ExactConversions.jl to start failing. Haven't looked into why, they probably just relied on the old behavior for some reason.

[JeffreySarnoff]

There are 3 lines in https://github.com/FedericoStra/ExactConversions.jl that use Base.exponent_max.
The first line in both versions of function maxcommon and the first line of function mincommon.
The current source uses (Base.exponent_max(T)-1) which corrects for the old nonconformant exponent_max.
By removing the -1 from each of the three expressions, their code ought not error with this PR.

[KristofferC]

Thanks for looking into it @JeffreySarnoff