Consider replacing some expression instanceof AST_Constant with expression.is_constant() across the codebase

[WofWca]

After creating #1141, I had an idea that it might not be the only case where such a change is beneficial.
I think mostly it would be good for when there is a constant with one (or more (#1142)) unary prefixes, which, I believe is pretty common because we usually transform true to !0 and undefined to void 0, which makes expression instanceof AST_Constant false.
The main issue here is that I don't quite understand why is_constant() returns false for AST_RegExp:

terser/lib/compress/evaluate.js

Line 109 in 540e886

return !(this instanceof AST_RegExp);

Here's the commit that introduced this: 0610c02
Might just try doing this anyway and just seeing if any tests fail?

And I'd be glad to work on this.

[jridgewell]

The main issue here is that I don't quite understand why is_constant() returns false for AST_RegExp:

is_constant is badly named, and it'd be better to think of it as is_primitive. Regexs are instances of the RegExp and have state (particularly lastIndex).

[fabiosantoscode]

Also is_constant doesn't return true for anything that has an identity.

Here "num" can be inlined safely into the return (provided it's the only use case):

const num = 123
export function something() {
  return num
}

And now something() === something() always holds true. But the same wouldn't be the case if "num" was a regex.

[fabiosantoscode]

is_primitive sounds great!