[llvm] [GlobalISel] Add G_INSERT_SUBVECTOR to computeNumSignBits (PR #220976)

[Matt Arsenault via llvm-commits]

================
@@ -1469,3 +1471,37 @@ unsigned llvm::SignBitsOps::rot(unsigned SrcSignBits, unsigned BitWidth,
     return SrcSignBits - Amt;
   return 1;
 }
+
+unsigned llvm::SignBitsOps::insertSubvector(
+    ElementCount SrcEC, ElementCount SubEC, uint64_t Idx,
+    const APInt &DemandedElts,
+    function_ref<unsigned(unsigned, const APInt &)> ComputeNumSignBits,
+    function_ref<unsigned(unsigned)> ComputeNumSignBitsAllElts) {
----------------
arsenm wrote:

I'm not really a fan of the 2 callback arguments, it almost makes factoring this out not worth it. I'd want to get it down to 1, maybe move to a template arg, or just give up on the abstraction 

https://github.com/llvm/llvm-project/pull/220976