[Lang] Matrix/Vector refactor: support basic matrix ops

[AD1024]

Related issue = #5478
Refactored & Combined PR: #5797 and #5861
A part of PR #5551

[netlify]

✅ Deploy Preview for docsite-preview ready!

Name	Link
🔨 Latest commit	486e6b3
🔍 Latest deploy log	https://app.netlify.com/sites/docsite-preview/deploys/632e88a1a9bf640008e1774e
😎 Deploy Preview	https://deploy-preview-6077--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]

1. Shall we add tests for the modifications?
2. Why both UnaryOp and BinaryOp are modified in codegen, but only BinaryOp is modified in frontend_ir?
3. Shall we have is_real_scalar(), is_real_tensor(), and a general is_real() so that we don't have to use is_real() || is_real_tensor() everywhere?

[jim19930609]

You can probably separate out the codegen part, and add some c++ tests to test the type promotion and broadcasting.

Some examples can be found in: tests/cpp/ir/ir_type_promotion_test.cpp and tests/cpp/ir/frontend_type_inference_test.cpp

[AD1024]

1. Shall we add tests for the modifications?
2. Why both UnaryOp and BinaryOp are modified in codegen, but only BinaryOp is modified in frontend_ir?
3. Shall we have is_real_scalar(), is_real_tensor(), and a general is_real() so that we don't have to use is_real() || is_real_tensor() everywhere?

1. Sure! will add some tests to cover c++ side changes.
2. The reason is we are only covering elementwise operations now at codegen. LLVM has corresponding variant of the common binary ops (+ - * /, etc.) for vector types so we don't need to change to. For UnaryOps, the major change is on casting because we need to cast every individual elements in a VectorType if a tensor is being cast to another tensor type.
3. In the first version of the implementation I put them together but this change definitely caused some issue in other parts of the codebase. I tend to leave them as separate implementations.

[jim19930609]

In codegen_llvm.cpp, for those LLVM operations that supports both Primitive and TensorType operands, let's simplify the condition from sth like is_real_tensor(stmt->ret_type) || is_real(ret_type) to is_real(ret_type.get_element_type()), so it's agnostic of whether it's TensorType or not.