feat(llvm): Major refactor - BuilderCursor全域化 & Resolver API導入

Added:
- Resolver API (resolve_i64) for unified value resolution with per-block cache
- llvmlite harness (Python) for rapid PHI/SSA verification
- Comprehensive LLVM documentation suite:
  - LLVM_LAYER_OVERVIEW.md: Overall architecture and invariants
  - RESOLVER_API.md: Value resolution strategy
  - LLVM_HARNESS.md: Python verification harness

Updated:
- BuilderCursor applied to ALL lowering paths (externcall/newbox/arrays/maps/call)
- localize_to_i64 for dominance safety in strings/compare/flow
- NYASH_LLVM_DUMP_ON_FAIL=1 for debug IR output

Key insight: LoopForm didn't cause problems, it just exposed existing design flaws:
- Scattered value resolution (now unified via Resolver)
- Inconsistent type conversion placement
- Ambiguous PHI wiring responsibilities

Next: Wire Resolver throughout, achieve sealed=ON green for dep_tree_min_string

src/backend/llvm/compiler/codegen/instructions/arith.rs

@@ -10,6 +10,7 @@ use super::builder_cursor::BuilderCursor;
 pub(in super::super) fn lower_compare<'ctx, 'b>(
     codegen: &CodegenContext<'ctx>,
     cursor: &mut BuilderCursor<'ctx, 'b>,
+    resolver: &mut super::Resolver<'ctx>,
     cur_bid: BasicBlockId,
     func: &MirFunction,
     vmap: &HashMap<ValueId, BasicValueEnum<'ctx>>,
@@ -104,9 +105,11 @@ pub(in super::super) fn lower_compare<'ctx, 'b>(
     }
     let out = if let (Some(_li0), Some(_ri0)) = (as_int(lv), as_int(rv)) {
         // Localize integer operands into current block to satisfy dominance
-        let mut li = super::flow::localize_to_i64(codegen, cursor, cur_bid, *lhs, bb_map, preds, block_end_values, vmap)
+        let mut li = resolver
+            .resolve_i64(codegen, cursor, cur_bid, *lhs, bb_map, preds, block_end_values, vmap)
             .unwrap_or_else(|_| as_int(lv).unwrap());
-        let mut ri = super::flow::localize_to_i64(codegen, cursor, cur_bid, *rhs, bb_map, preds, block_end_values, vmap)
+        let mut ri = resolver
+            .resolve_i64(codegen, cursor, cur_bid, *rhs, bb_map, preds, block_end_values, vmap)
             .unwrap_or_else(|_| as_int(rv).unwrap());
         // Normalize integer widths: extend the narrower to match the wider to satisfy LLVM
         let lw = li.get_type().get_bit_width();