Loopy is slow in make_kernel, preprocess_kernel and codegen

Thanks to @kaushikcfd, scheduling is now super fast compared to other parts of loopy. Still, `make_kernel, preprocess_kernel, codegen` take so much time that some sumpy kernels are unusable.

Here's a small example with https://github.com/isuruf/sumpy/tree/derivtaker

```python
import numpy as np
import sys
import loopy as lp

import pyopencl as cl

from sumpy.expansion.multipole import LaplaceConformingVolumeTaylorMultipoleExpansion
from sumpy.expansion.local import LaplaceConformingVolumeTaylorLocalExpansion
from sumpy.kernel import LaplaceKernel
import sumpy.symbolic as sym

import logging

logger = logging.getLogger(__name__)

try:
import faulthandler
except ImportError:
pass
else:
faulthandler.enable()

knl = LaplaceKernel(3)
local_expn_class = LaplaceConformingVolumeTaylorLocalExpansion
mpole_expn_class = LaplaceConformingVolumeTaylorMultipoleExpansion
order = 12
ctx_factory = cl._csc

logging.basicConfig(level=logging.INFO)

from sympy.core.cache import clear_cache

clear_cache()

ctx = ctx_factory()
queue = cl.CommandQueue(ctx, properties=cl.command_queue_properties.PROFILING_ENABLE)

np.random.seed(17)

target_kernels = [knl]

m_expn = mpole_expn_class(knl, order=order)
l_expn = local_expn_class(knl, order=order)

from sumpy import P2EFromSingleBox, E2PFromSingleBox, P2P, E2EFromCSR

m2l = E2EFromCSR(ctx, m_expn, l_expn)

loopy_knl = m2l.get_optimized_kernel()
loopy_knl = lp.add_and_infer_dtypes(
loopy_knl,
dict(
tgt_ibox=np.int32,
centers=np.float64,
tgt_center=np.float64,
target_boxes=np.int32,
src_ibox=np.int32,
src_expansions=np.float64,
tgt_rscale=np.float64,
src_rscale=np.float64,
src_box_starts=np.int32,
src_box_lists=np.int32,
),
)
lp.generate_code_v2(loopy_knl)
```