[mlir][tosa] Fix tosa.concat by inserting linalg.fill after linalg.init

[rsuderman]

Summary

All linalg.init operations must be fed into a linalg operation before
subtensor. The inserted linalg.fill guarantees it executes correctly.

[rsuderman]

Fixed tosa-to-linalg-on-tensors test

mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp

matchAndRewrite(tosa::ConcatOp op, ArrayRef<Value> args,
 
     Value resultDimSize = sizes[axis];
     for (auto arg : args.drop_front()) {
       auto size = rewriter.create<memref::DimOp>(loc, arg, axisValue);
       resultDimSize = rewriter.create<AddIOp>(loc, resultDimSize, size);
     }
     sizes[axis] = resultDimSize;
 
-    Value result = rewriter.create<linalg::InitTensorOp>(
+    Value init = rewriter.create<linalg::InitTensorOp>(
         loc, resultType.getShape(), resultType.getElementType());
 
+    Value zeroVal = rewriter.create<ConstantOp>(
+        loc, rewriter.getZeroAttr(resultType.getElementType()));
+    Value result =
+        rewriter.create<linalg::FillOp>(loc, init, zeroVal).getResult(0);
+
     for (auto arg : args) {
       sizes[axis] = rewriter.create<memref::DimOp>(loc, arg, axisValue);
       result = rewriter.create<SubTensorInsertOp>(loc, arg, result, offsets,
                                                   sizes, strides);
       offsets[axis] = rewriter.create<AddIOp>(loc, offsets[axis], sizes[axis]);
     }
     rewriter.replaceOp(op, result);
     return success();