Add arange op to improve readibility of creating multdimensional tensors

[Hprairie]

Follow a similar API to einx. However, we also want to add new functionality to specify start, stop, and stride.

First a note, until accepting kwargs is allowed, we need to use tlib's built in dict object. I want to find a way to shorten this down, but it is fine for now. Here is the reference API:

// With only a stop (we assume a_stride = b_end)

tlib.arange("a b", tlib.dict(a=2, b=3))

# Will create a 5 by 10 arange, that goes 

# 0, 1, 2
# 3, 4, 5

// With a stop and stride

tlib.arange("a b", tlib.dict(a=(2, 10), b=(3, 1))

# Will create a 5 by 10 arange, that goes 

# 0, 1, 2
# 10, 11, 12

I am still trying to decide if setting a start is really necessary. We only technically need to do it once. Need to think it through.

[Hprairie]

After more deliberation, I am leaning towards stop and (stop, stride). As the possible options, we can make the following code block:

    offs_am = (pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)) % M
    offs_bn = (pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)) % N
    offs_k = tl.arange(0, BLOCK_SIZE_K)
    a_ptrs = a_ptr + (offs_am[:, None] * stride_am + offs_k[None, :] * stride_ak)
    b_ptrs = b_ptr + (offs_k[:, None] * stride_bk + offs_bn[None, :] * stride_bn)

into

    offs_m = pid_m * BLOCK_SIZE_M
    offs_n = pid_n * BLOCK_SIZE_N
    a_ptrs = a_ptr + offs_m + tlib.arange("m k", tlib.dict(m=(BLOCK_SIZE_M, stide_am), k=(BLOCK_SIZE_K, stride_ak))
    b_ptrs = b_ptr + offs_N + tlib.arange("k n", tlib.dict(n=(BLOCK_SIZE_N, stide_an), k=(BLOCK_SIZE_K, stride_ak))

With this, the idea is that you really only need to read the string to understand the layout.

[Hprairie]

tlib.arange has now been implemented, however, I am wanting to add logic to auto determine masking. I can technically keep adding more stuff into the tuple, however, this could make it rather confusing. I am wondering how to best approach this.