Scheduling stage is slow for many-instruction kernel with disjoint loop nests

Reproducer:
```python
import loopy as lp

def main():
n = 500
insns_as_str = [f"y{i}[idim{i},jdim{i}] = x{i}[idim{i},jdim{i}]"
for i in range(n)]

domains = [f"{{[idim{i},jdim{i}]: 0<=idim{i},jdim{i}<2}}"
for i in range(n)]

x_args = [lp.GlobalArg(f"x{i}", shape=lp.auto, dtype=float)
for i in range(n)]

knl = lp.make_kernel(domains,
"\n".join(insns_as_str),
x_args + [...],
lang_version=(2018, 2))

print(lp.generate_code_v2(knl).device_code())

if __name__ == "__main__":
main()
```