[bugfix] Properly detect kwargs hashes vs optional positional args

[malcolmohare]

Attempting to fix https://github.com/rspec/rspec-expectations/issues/1451

Refactor has_kw_args_in?:

• try to use positional parameter information to determine whether a hash is kw_args
• try to leverage information passed by the ruby2_keywords annotation
• fallback to inspection of keyword args to determine if they look valid or not

Added backwards compatible ruby2_keywords wrapping to the necessary methods (valid? and error_for) in the spec files like so:
ruby2_keywords(:valid?) if respond_to?(:ruby2_keywords, true)

Also needed to modify how args are passed to the MSV split_args so that the special kwargs markings are not lost.

[malcolmohare]

Fully backwards compatible, just need to fix 3.2 and up. The behavior of ruby2_keywords changed between 3.1 and 3.2: https://rubyreferences.github.io/rubychanges/3.2.html#keyword-argument-separation-leftovers

[malcolmohare]

something had to change here. Splatting the args meant the special marking for the kwargs hash gets lost. This doesnt happen to functions with the ruby2_keywords... except it won't apply here because this function doesn't take the *args input. So either the args array is cloned (which persists the special flagging) or the input to this function needs to swap to using *args. I figured cloning was the less obtrusive change. The split_args function is private so changing its signature was not something that has compatibility concerns.

[JonRowe]

Apologies I never got round to looking at this, I've migrated it over to the monorepo as rspec/rspec#121 I'm hoping to do a big refactoring of keyword argument support with RSpec 4 dropping older ruby support.