[Feature] Overwriting a layout and then building results in a confusing error message

[dl2239]

Motivation and Expected Behavior

In my code I have two declarations of linear_C_layout, each assigned a different layout. When I built the kernel that the layout was in, I got this error message: `IndexError: tuple index out of range

Error: tuple index out of range`. I believe this error message is confusing and have contacted @Fangtangtang about it.

Related resources

No response

[Fangtangtang]

to reproduce

from allo.ir.types import float32
import allo.dataflow as df
from allo.memory import Layout

S = Layout.Shard
R = Layout.Replicate

Ty = float32

linear_A_layout = [S(0), R]
linear_C_layout = [R, S(0)] # ⚠️ first assignment to linear_C_layout 

A_x = 4
C_x = 1

@df.region()
def copy(A: Ty[A_x, 768], C: Ty[C_x, 768*4]):
    @df.kernel(mapping=[4], args=[A, C])
    def mod(
        local_A: Ty[A_x, 768] @ linear_A_layout,
        local_C: Ty[C_x, 768*4] @ linear_C_layout,
    ):
        local_C[:,:] = local_A[:,:]

linear_A_layout = [S(0), R]
linear_C_layout = [S(0), S(1)] # ⚠️ second assignment to linear_C_layout 

copy_mod = df.build(copy, target="aie", project="copy.prj")

The root cause is that global variables are captured during the build phase (call df.build), so the initial assignment to linear_C_layout gets overwritten by the subsequent one.