[FIX][LLVM] Support bool Add/Sub/Mul in LLVM codegen

[wwoosshh]

Fixes #20558.

+, - and * on bool operands hit Check failed: (t.MatchesCode(DLDataTypeCode::kDLFloat)) in the LLVM codegen, because DEFINE_CODEGEN_BINARY_OP only handled int, uint and float. The C target builds the same function.

This PR follows the C target. The bool operands are zero-extended to i32, the operation is done in i32, and the result is converted back to bool by comparing it with zero. For 0/1 inputs this gives OR for +, XOR for - and AND for *, and in the optimized IR (inspect_source("ll")) the vectorized case reduces to exactly or, xor and and. The same code path covers vector bool types.

To check that the behavior matches, I built the same PrimFunc for target="c", compiled it with export_library, and compared the stored bytes with the LLVM build for all four input combinations. They are identical.

In the issue I also asked whether bool arithmetic should instead be rejected earlier. I can switch to that approach if you prefer it.

Testing: added test_llvm_bool_arith to tests/python/codegen/test_target_codegen_llvm.py. It covers all input combinations for +, - and * with both a serial and a vectorized loop. It fails on main with the check above and passes with this change. The full test_target_codegen_llvm.py passes locally (381 passed).

Generated-by: Claude Code (Claude Opus 5.5)

[wwoosshh]

cc @tlopex @tqchen

This is a small fix for #20558: the LLVM codegen now handles bool +/-/* the same way as the C target (compute in int, then non-zero is true), with a regression test. Could you take a look when you have time? Thanks!