windows: remove the 8192-codepoint arg limit in FuzzComposeCommandLine

It just occurred to me that the observed limit was almost certainly a
side effect of the Go wrapper for CommandLineToArgv (golang/go#63236)
rather than a behavior of the system call itself.

Updates golang/go#63236.
Updates golang/go#58817.

Change-Id: Icc9db01f201f54a78044d1c48e0883e098cfb5e5
Cq-Include-Trybots: luci.golang.try:x_sys-gotip-windows-amd64-longtest
Reviewed-on: https://go-review.googlesource.com/c/sys/+/531176
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: Bryan Mills <bcmills@google.com>
Reviewed-by: Quim Muntal <quimmuntal@gmail.com>
Reviewed-by: Than McIntosh <thanm@google.com>

Author: Bryan C. Mills

windows/syscall_windows_test.go

@@ -589,6 +589,7 @@ func FuzzComposeCommandLine(f *testing.F) {
 	f.Add(`C:\"Program Files"\Go\bin\go.exe` + "\x00env")
 	f.Add(`C:\"Pro"gram Files\Go\bin\go.exe` + "\x00env")
 	f.Add("\x00" + strings.Repeat("a", 8192))
+	f.Add("\x00" + strings.Repeat("a", 8193))
 	f.Add(strings.Repeat("\x00"+strings.Repeat("a", 8192), 4))
 
 	f.Fuzz(func(t *testing.T, s string) {
@@ -629,16 +630,6 @@ func FuzzComposeCommandLine(f *testing.F) {
 					// have one or more runes converted to replacement characters.
 					t.Skipf("skipping: input %d is not valid UTF-8", i)
 				}
-				if len(arg) > 8192 {
-					// CommandLineToArgvW seems to truncate each argument after 8192
-					// UTF-16 code units, although this behavior is not documented. Since
-					// it isn't documented, we shouldn't rely on it one way or the other,
-					// so skip the input to tell the fuzzer to try a different approach.
-					enc, _ := windows.UTF16FromString(arg)
-					if len(enc) > 8192 {
-						t.Skipf("skipping: input %d encodes to more than 8192 UTF-16 code units", i)
-					}
-				}
 			}
 			if testing.Verbose() {
 				t.Logf("using input: %#q", args)