move copy kernel to out of schedule ordering

[Qazalin]

@geohot I think the scheduler should create the correct AST in the DAG transformation part.

[github-actions]

Changes

Name                           Lines    Diff    Tokens/Line    Diff
---------------------------  -------  ------  -------------  ------
tinygrad/engine/schedule.py      230      +0           14.6    +0.0


total lines changes: 0

[Qazalin]

ran this with USE_COPY_KERNEL=1

import numpy as np
from tinygrad import Tensor
from tinygrad.engine.realize import run_schedule
from tinygrad.engine.schedule import create_schedule

a = Tensor.randn(4, 4, device="HSA:1").realize()
c = Tensor.ones((4, 4), device="HSA:1").contiguous().realize()
b = a.lazydata._copy("HSA:2")
#b = c.lazydata._copy("HSA:2")
sched = create_schedule([b])
run_schedule(sched.copy())
np.testing.assert_equal(np.frombuffer(a.lazydata.buffer.as_buffer(), a.dtype.np), np.frombuffer(b.buffer.as_buffer(), b.dtype.np))

to test. Would be great if we can have CI.