NodePattern: Optimize by using placeholders for post processing

[marcandre]

The improved handling of ... had to use some post-processing of the cur_node (#6843)
The easiest way to provide other "multiple match" patterns (e.g. #6841) would involve similar post-processing. This PR paves the way by using post processing of the current node pattern everywhere.

This as other multiple benefits:

• deals automatically and better with generated temporary variables
• much less argument passing to the various compiling methods
• factorizes the #{'.type' if seq_head} pattern that was everywhere.
• makes debugging easier with simpler & nicer generated code

I had hoped to be able to minimize the number of node guards, but unions make it tricky.

For the record, here's an example of the code generated before & after.

### Pattern:
($_ {:a :b} nil?)

### Before:
(temp1 = node0; temp1 = temp1;temp1.is_a?(RuboCop::AST::Node) &&
(capture1 = temp1.type; true) &&
(temp2 = temp1.children[0]; temp2 = temp2;(temp2 == :a) || (temp2 == :b)) &&
(temp1.children[1].nil?) &&
(temp1.children.size == 2))

### After:
node0.is_a?(RuboCop::AST::Node) &&
(capture1 = node0.type; true) &&
((temp1 = node0.children[0]) == :a || temp1 == :b) &&
node0.children[1].nil? &&
node0.children.size == 2

[bbatsov]

Fantastic work!