Fix nested for loops: emit inner while(true) for nested loop headers

The bug: emit_loop_body! never checked for nested loop headers.
Inner loops compiled as if-blocks instead of while(true) loops,
causing inner break to exit the outer loop.

Fix: when emit_loop_body! encounters a statement that's in loop_headers
(and isn't the current loop's own header), emit a nested while(true)
with proper phi initialization and recursive body emission.

Now works: double nested, triple nested, matmul with ND arrays.

Tests: 3 new e2e Node.js tests
- nested loop sum (1:3 × 1:2 = 18)
- triple nested counter (2×2×2 = 8)
- 2×2 matrix multiply via triple nested loop + ND arrays = [19,43,22,50]

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: Dale-Black

src/compiler/codegen.jl

@@ -337,6 +337,43 @@ function emit_loop_body!(ctx, buf, code, start_idx, end_idx, loop_headers, backw
     while i <= end_idx
         stmt = code[i]
 
+        # ─── Nested loop header: emit inner while(true) ───
+        if i in loop_headers && i != loop_header
+            inner_loop_end = 0
+            for (src, tgt) in backward_edges
+                if tgt == i
+                    inner_loop_end = src
+                    break
+                end
+            end
+            if inner_loop_end > 0
+                # Initialize inner loop phi variables
+                for phi_idx in sort(collect(keys(phi_info)))
+                    if phi_idx >= i && phi_idx <= inner_loop_end
+                        phi = phi_info[phi_idx]
+                        var_name = ctx.js_locals[phi_idx]
+                        for (k, edge) in enumerate(phi.edges)
+                            if edge < i
+                                if isassigned(phi.values, k)
+                                    val = compile_value(ctx, phi.values[k])
+                                    print(buf, "$(indent)$(var_name) = $(val);\n")
+                                end
+                            end
+                        end
+                    end
+                end
+                print(buf, "$(indent)while (true) {\n")
+                inner_body_start = i
+                while inner_body_start <= inner_loop_end && code[inner_body_start] isa Core.PhiNode
+                    inner_body_start += 1
+                end
+                emit_loop_body!(ctx, buf, code, inner_body_start, inner_loop_end, loop_headers, backward_edges, forward_gotos, phi_info, depth + 1, i)
+                print(buf, "$(indent)}\n")
+                i = inner_loop_end + 1
+                continue
+            end
+        end
+
         if stmt isa Core.GotoIfNot
             cond = compile_value(ctx, stmt.cond)
             target = stmt.dest

test/runtests.jl

@@ -3551,7 +3551,34 @@ process.stdout.write(JSON.stringify(r));
         end
         @test compile_unopt_and_run(fn, "") == "[19,43,22,50]"
     end
-    # Note: nested for loops (for i; for j; for k) have a known JST bug
-    # where inner loop `break` exits the outer while loop. Single for loops work.
-    # TODO: fix nested for loop compilation in JST
+    @testset "ND: nested loop sum" begin
+        fn = () -> begin
+            total = 0
+            for i in 1:3; for j in 1:2; total = total + i * j; end; end
+            return total
+        end
+        @test compile_unopt_and_run(fn, "") == "18"
+    end
+    @testset "ND: triple nested loop" begin
+        fn = () -> begin
+            total = 0
+            for i in 1:2; for j in 1:2; for k in 1:2; total = total + 1; end; end; end
+            return total
+        end
+        @test compile_unopt_and_run(fn, "") == "8"
+    end
+    @testset "ND: matmul 2x2 (triple nested)" begin
+        fn = () -> begin
+            A=zeros(2,2); A[1,1]=1.0; A[1,2]=2.0; A[2,1]=3.0; A[2,2]=4.0
+            B=zeros(2,2); B[1,1]=5.0; B[1,2]=6.0; B[2,1]=7.0; B[2,2]=8.0
+            C=zeros(2,2)
+            for i in 1:2; for j in 1:2; s=0.0
+                for k in 1:2; s=s+A[i,k]*B[k,j]; end
+                C[i,j]=s
+            end; end
+            C
+        end
+        # A*B = [[19,22],[43,50]], column-major flat = [19,43,22,50]
+        @test compile_unopt_and_run(fn, "") == "[19,43,22,50]"
+    end
 end