Numeric#to_int is missing: SystemCallError.new with a Complex errno raises TypeError where MRI truncates, or raises RangeError #631

MRI's NUM2LONG reaches a non-Integer through rb_to_int, i.e. #to_int. Numeric#to_int is defined as to_i (numeric.c num_to_int), so every Numeric answers it:

$ ruby -e 'p Complex(2.9, 0).to_int'          #=> 2
$ ruby -e 'p Complex(2.9, 1).to_int'          #=> RangeError: can't convert 2.9+1i into Integer
$ ruby -e 'p Complex(2.9, 0).respond_to?(:to_int)'  #=> true
$ ruby -e 'p Complex.instance_methods(false).grep(/to_i/)'  #=> [:to_i]

rbgo has Complex#to_i with the right value and the right RangeError, but no to_int, because Numeric#to_int is not defined:

$ rbgo -e 'p Complex(2.9, 0).to_i'    #=> 2
$ rbgo -e 'p Complex(2.9, 1).to_i'    #=> RangeError: can't convert (2.9+1i) into Integer
$ rbgo -e 'p Complex(2.9, 0).to_int'  #=> NoMethodError: undefined method 'to_int' for an instance of Complex

Consequence, now that SystemCallError#initialize coerces its errno the way syserr_initialize does (#629):

• SystemCallError.new('foo', Complex(2.9, 0)) is a TypeError where MRI builds Errno::ENOENT
• SystemCallError.new('foo', Complex(2.9, 1)) is a TypeError where MRI raises RangeError: can't convert 2.9+1i into Integer

Cost: 2 examples in core/exception/system_call_error_spec.rb ("converts to Integer if errno is a Complex convertible to Integer", "raises RangeError if errno is a Complex not convertible to Integer"). Anything else that coerces through #to_int is affected the same way.

Fix belongs in Numeric (one alias of to_i), not at the call sites — the #629 branch deliberately did not work around it locally, because a second coercion rule would drift from the first.

Note the message also differs: MRI says can't convert 2.9+1i into Integer, rbgo can't convert (2.9+1i) into Integer (parenthesised).