new gensym handling

[Araq]

• will fix many template bugs, eventually.

[timotheecour]

nit: is optNimV019 related to nimv019 in config/nim.cfg? as i understand this only affects gensym handling, so maybe it should be in the flag name, eg --genSymVersion, as there could be other transition flags that should be settable independently; eg if in future you'll need a transition between 1.0 and 1.1 that's totally unrelated to gensym handling

[Araq]

We are moving over to "trunk based" development, that's why. These transition switches will be collected under useVersion.

[timotheecour]

I thought this PR would fix following bugs but apparently it doesn't, and maybe it can't as there's inherent ambiguity:

type Foo = object
  b: int
proc bar(b = 12) = discard

template fun(b): untyped =
  var x: Foo
  # each line here would give an error
  var x0 = Foo(b: 3)
  bar(b=32)
  x.b = 5
  echo x.b

template fun2(b2): untyped =
  var x = Foo(b: 3)
  x.b = 5
  echo x.b
  echo b2

proc main()=
  fun2(14) # ok
  fun(14) # Error: identifier expected, but found '14'

main()

IMO we should reopen #8899, it's a fool proof solution:

template fun(b): untyped =
  var x: Foo
  var x0 = Foo(`b`: 3)
  bar(`b`=32)
  x.`b` = 5
  x.b = 15 # this one actually means: substitute (eg if user passes an identifier as b, it will work as intended)
  echo x.`b`

since there's no need to quote valid identifiers (such as template arguments), there's no loss of generality.
more examples showing user can still pass a quoted ident as argument:

template baz(field) =
  field(32)
baz(`x=`) # will call `x=`(32)
baz(bar) # will call bar(32)