Generalize strides

**Pros**:

`no-op` in the following statements refers to no kernel launch. it would require some changes to meta-data.
- `transpose` and `reorder` are now no-ops (we can have an additional parameter if we want to force copies).
- All indexing is now a no-op.
- `wrap` and `unwrap` can be no-ops in a few cases and would not require additional memory.
- `tile` can be a no-op without additional memory.

**Cons**:
- Most of kernels are based on the assumption that the first dimension is contiguous, would require a lot of code changes.
- Even if we change kernels, it might create performance drops.
- Creating new bugs.