[stdlib] Improve InlineArray move/copy for trivial types.

The `InlineArray` copy/move constructors current implementation simply
looped through each element and moved/copied. This produced poor codegen
as it does not take advantage of any loop unrolling or checking if the
element types are trivial and simply copying the underlying MLIR
storage.

The previous codegen for a simple function:
```mojo
fn return_array() -> InlineArray[Int32, 4]:
    var arr = InlineArray[Int32, 4](fill=0)
    return arr^
```

produces this codegen:
```
define dso_local void @"test_inline_array::return_array"(ptr noalias noundef nonnull writeonly captures(none) %0) #0 !dbg !5 {
  tail call void @llvm.memset.p0.i64(ptr noundef nonnull align 4 dereferenceable(16) %0, i8 0, i64 16, i1 false), !dbg !90
  tail call void asm sideeffect "nop", ""() #2, !dbg !91
  ret void, !dbg !32
}
```

but now, we get this improved LLVM IR:
```
define dso_local void @"test_inline_array::_return_array"(ptr noalias noundef nonnull writeonly captures(none) initializes((0, 16)) %0) #0 !dbg !5 {
  tail call void @llvm.memset.p0.i64(ptr noundef nonnull align 4 dereferenceable(16) %0, i8 0, i64 16, i1 false), !dbg !51
  ret void, !dbg !32
}
```

Notably there is no `asm sideffect "nop"` and an `initializes((0, 16))`
is on the `ptr` argument showing that the values do get initialized.

MODULAR_ORIG_COMMIT_REV_ID: 14efb85038ec4697b7e52a8b7beb76e2b48b09bc

Author: NathanSWard

mojo/stdlib/std/collections/inline_array.mojo

@@ -352,11 +352,14 @@ struct InlineArray[ElementType: Copyable, size: Int](
         ```
         """
 
-        self = Self(uninitialized=True)
-
-        for idx in range(Self.size):
-            var ptr = self.unsafe_ptr() + idx
-            ptr.init_pointee_copy(other.unsafe_get(idx))
+        @parameter
+        if Self.ElementType.__copyinit__is_trivial:
+            self._array = other._array
+        else:
+            self = Self(uninitialized=True)
+            for idx in range(Self.size):
+                var ptr = self.unsafe_ptr() + idx
+                ptr.init_pointee_copy(other.unsafe_get(idx))
 
     fn __moveinit__(out self, deinit other: Self):
         """Move constructs the array from another array.
@@ -368,11 +371,14 @@ struct InlineArray[ElementType: Copyable, size: Int](
             Moves the elements from the source array into this array.
         """
 
-        __mlir_op.`lit.ownership.mark_initialized`(__get_mvalue_as_litref(self))
-
-        for idx in range(Self.size):
-            var other_ptr = other.unsafe_ptr() + idx
-            (self.unsafe_ptr() + idx).init_pointee_move_from(other_ptr)
+        @parameter
+        if Self.ElementType.__moveinit__is_trivial:
+            self._array = other._array
+        else:
+            self = Self(uninitialized=True)
+            for idx in range(Self.size):
+                var other_ptr = other.unsafe_ptr() + idx
+                (self.unsafe_ptr() + idx).init_pointee_move_from(other_ptr)
 
     fn __del__(deinit self):
         """Deallocates the array and destroys its elements."""

mojo/stdlib/test/collections/test_inline_array.mojo

@@ -13,6 +13,7 @@
 
 from sys.info import size_of
 
+from compile import compile_info
 from memory.maybe_uninitialized import UnsafeMaybeUninitialized
 from test_utils import CopyCounter, DelRecorder, MoveCounter, check_write_to
 from testing import assert_equal, assert_true, assert_false, TestSuite
@@ -392,5 +393,30 @@ def test_inline_array_triviality():
     assert_false(InlineArray[String, 1].__moveinit__is_trivial)
 
 
+fn _return_array[copy: Bool = False]() -> InlineArray[Int32, 4]:
+    var arr = InlineArray[Int32, 4](fill=0)
+
+    @parameter
+    if copy:
+        return arr.copy()
+    else:
+        return arr^
+
+
+def test_inline_array_copy_and_move_llvm_ir():
+    def _test(ir: StringSlice):
+        assert_true("initializes((0, 16))" in ir)
+        assert_false('asm sideeffect "nop"' in ir)
+
+    var move_info = compile_info[
+        _return_array[copy=False], emission_kind="llvm-opt"
+    ]()
+    _test(move_info.asm)
+    var copy_info = compile_info[
+        _return_array[copy=True], emission_kind="llvm-opt"
+    ]()
+    _test(copy_info.asm)
+
+
 def main():
     TestSuite.discover_tests[__functions_in_module()]().run()