Fix the scope and argument binding of a lambda literal

[sinsoku]

Motivation / Background

This Pull Request has been created because reviewing #481 after it merged turned up five defects:

def foo
  x = 1
  ->(x) { }.call("str")
  x                             #=> Integer, inferred as String
end

->(h) { h }.call(k: 1)          #=> wrong number of arguments (0 for 1)
->(&b) { b }.call { 1 }         #=> untyped, not Proc
[1, 2].map(&->(*a) { a })       #=> Array[Array[untyped]], not Array[Array[Integer]]
[1].each { |x: helper(1)| x }   #=> no diagnostic for the call in the default

Detail

One commit per defect, in the order above.

• An empty body built its DummyNilNode with the enclosing LocalEnv, which install0 then takes as the block's own scope.
• Keywords passed to formals that take none are the last positional hash, which only the method path did before binding.
• Proc#call receives the block it is given, but nothing bound it to &b.
• A lambda passed with & still went through the block-shaped binder, which knows only positionals and checks no arity.
• Only LambdaNode listed keyword defaults as subnodes, so in a block nothing installed the nodes AST.parse_params had built.

A block still does not bind keywords; only the default expression is analyzed. lambda {}, proc {} and Proc.new {} lose Type::Proc through their RBS declarations, so none of the argument binding reaches them.

The last commit fixes nothing: it covers the block of super() { ... }, which #481 made report a return type mismatch instead of printing the node class from wrong_return_type.

Verification

rake test passes at every commit, and each added scenario, except the one for super, fails on the commit before its fix.

🤖 Generated with Claude Code