[Lang] MatrixNdarray refactor part4: Lowered TensorType to CHI IR level for elementwise-indexed MatrixNdarray

[jim19930609]

Related issue = #5873, #5819

This PR is working on "Part ②", "Part ③" and "Part ④" in #5873.

[Overview]
When "real_matrix = True", suppose we have the following code with elementwise-indexed MatrixNdarray involved:

@ti.kernel
def func(a: ti.types.ndarray()):
    for i in range(5):
        for j in range(4):
            a[i][j * j] = j * j 

v = ti.Vector.ndarray(10, ti.i32, 5, layout=ti.AOS)
func(v)

The lowering pass will be:

[Major changes]

1. Change on Frontend Language
   In python/taichi/lang/impl.py, we modified subscript methods to turn-off scalarization for ExternalTensorExpression.

2. Change on IR (with real_matrix = True):

• ArgLoadStmt: Now able to return TensorType
• PtrOffsetStmt: Now able to take ExternalPtrStmt as base pointer
• IndexExpression: Now able to index ExternalTensorExpression

3. Change on CodeGen
   Adjusted codegen logic for ArgLoadStmt, PtrOffsetStmt and ExternalPtrStmt to correctly handle TensorType

[netlify]

✅ Deploy Preview for docsite-preview ready!

Name	Link
🔨 Latest commit	18cecc2
🔍 Latest deploy log	https://app.netlify.com/sites/docsite-preview/deploys/631195a961c4580008fe7cda
😎 Deploy Preview	https://deploy-preview-5936--docsite-preview.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site settings.

[strongoier]

Great!! Only some nits.