compute::generic apply with inferred type args 1m

Exercises inferred-generic-call frame seeding (a LoadType + wider Call per iteration): gapply<T>'s T is inferred at every call site, so the caller seeds frame.type_args = [int] each time. Guards the cost of CallPlan.type_args threading against regressions; the rest of the compute suite is monomorphic and cannot see it.

BAML

baml

function gapply<T>(f: (T) -> T, x: T) -> T { f(x) }
function main() -> int {
  let double = (x: int) -> int { x * 2 };
  let s = 0;
  for (let i = 0; i < 1000000; i += 1) { s += gapply(double, i); };
  return s;
}

Python

python

def gapply(f, x): return f(x)
double = lambda x: x * 2
s = 0
for i in range(1000000): s += gapply(double, i)
print(s)

TypeScript

function gapply(f,x){return f(x)}
const double=x=>x*2;let s=0;
for(let i=0;i<1000000;i++)s+=gapply(double,i);console.log(s)