abs for float

[paugier]

abs(1.2) gives

 Traceback (most recent call last):
  * bug_abs::main at ~/dev/spy-demos/loop-fusion/bug_abs.spy:4
  |     print(abs(1.2))
  |           |______|

TypeError: mismatched types
  | ~/dev/spy-demos/loop-fusion/bug_abs.spy:4
  |     print(abs(1.2))
  |               |_| expected `i32`, got `f64`

  | ~/dev/spy/spy/vm/modules/builtins.py:34
  | def w_abs(vm: "SPyVM", w_x: W_I32) -> W_I32:
  | |__________________________________________| function defined here

[antocuni]

Agreed that it should be solved.
abs (but also e.g. min and max) is currently implemented at interp-level in builtins.py:

spy/spy/vm/modules/builtins.py

Lines 33 to 37 in 5ecdd8c

	@BUILTINS.builtin_func
	def w_abs(vm: "SPyVM", w_x: W_I32) -> W_I32:
	x = vm.unwrap_i32(w_x)
	res = vm.ll.call("spy_builtins$abs", x)
	return vm.wrap(res)

however, this is for historical reasons.
I think that nowadays it makes more sense to implement it in pure spy, so we can easily use generics and metafunctions to automatically dispatch on the right type.
Something like this:

@blue.metafunc
def abs(m_x):
    T = m_x.static_type
    return _abs[T]

def _abs[T](x: T) -> T:
    if x < 0:
        return -x
    return x

This should probably live in e.g. stdlib/_builtins.py, and consequently we should remove spy_builtin$abs from libspy