[mlir]: Remove arith/cf converion patterns from FuncToLLVM.cpp

[j2kun]

Creating this issue to reflect an existing TODO at

llvm-project/mlir/lib/Conversion/FuncToLLVM/FuncToLLVM.cpp

Line 807 in 09f1aac

// TODO: Remove these in favor of their dedicated conversion passes.

[j2kun]

Linked the TODO to this issue in #70984

[j2kun]

I started on this in j2kun#5, and many tests seem to rely on this in a way I don't yet understand. I'm getting lots of test failures complaining about unrealized conversion casts between index and i64