Add specs for Kernel.Float with hex strings

[herwinw]

They are very much driven by the current (inconsistent) behaviour of MRI. I'm not sure if these should be seen as a specification or as MRI specific implementation quirks.

[herwinw]

https://bugs.ruby-lang.org/issues/21163 is a related upstream case where I pointed out the inconsistencies.

[herwinw]

I've updated the PR to reflect the changes in Ruby 3.4.3. It now contains checks like this:

ruby_version_is ""..."3.4.3" do
  it "does not accept embedded _ if the number contains a-f" do
    -> { @object.send(:Float, "0x1_0a") }.should raise_error(ArgumentError)
  end
end

ruby_version_is "3.4.3" do
  it "accepts embedded _ if the number contains a-f" do
    @object.send(:Float, "0x1_0a").should == 0x10a.to_f
  end
end

I'm not sure if this should be checked with a version guard, or with a ruby_bug. It's not the kind of fix I would expect in a patchlevel release.

[eregon]

I'm not sure if this should be checked with a version guard, or with a ruby_bug. It's not the kind of fix I would expect in a patchlevel release.

If not likely to be backported then yes a version guard is correct.

I'm not sure if these should be seen as a specification or as MRI specific implementation quirks.

Regardless it's good to test it, and indeed report weird behavior upstream.
If something looks strange in the specs adding a code comment with a link to the issue is helpful.

[eregon]

Ah actually it got backport, but only to 3.4.3. Yeah, what you did is right.