Build zstd with multithreading enabled (#762)

I verified this on macOS

```
❯ python -c "from compression import zstd; print(zstd.CompressionParameter.nb_workers.bounds())"
(0, 256)
```

See #761 

I'm not sure if this is all of the performance difference, but using
threads should make a significant difference.
#761 (comment)
notes that threads are not used by default, so I think there's more
going on — but we should enable this at build time regardless. See also
#768.

Author: zanieb

cpython-unix/build-zstd.sh

@@ -57,7 +57,7 @@ index 5e6e8bc..6ca72a1 100644
 EOF
 fi
 
-CFLAGS="${EXTRA_TARGET_CFLAGS} -fPIC" LDFLAGS="${EXTRA_TARGET_LDFLAGS}" make -j ${NUM_CPUS} libzstd.a
+CFLAGS="${EXTRA_TARGET_CFLAGS} -DZSTD_MULTITHREAD -fPIC" LDFLAGS="${EXTRA_TARGET_LDFLAGS}" make -j ${NUM_CPUS} libzstd.a
 make -j ${NUM_CPUS} install-static DESTDIR=${ROOT}/out
 make -j ${NUM_CPUS} install-includes DESTDIR=${ROOT}/out
-make -j ${NUM_CPUS} install-pc DESTDIR=${ROOT}/out
+MT=1 make -j ${NUM_CPUS} install-pc DESTDIR=${ROOT}/out

src/verify_distribution.py

@@ -172,6 +172,15 @@ def test_gil_disabled(self):
 
         self.assertEqual(sysconfig.get_config_var("Py_GIL_DISABLED"), wanted)
 
+    @unittest.skipIf(
+        sys.version_info[:2] < (3, 14),
+        "zstd is only available in 3.14+",
+    )
+    def test_zstd_multithreaded(self):
+        from compression import zstd
+
+        assert zstd.CompressionParameter.nb_workers.bounds() == (0, 256)
+
     @unittest.skipIf("TCL_LIBRARY" not in os.environ, "TCL_LIBRARY not set")
     @unittest.skipIf("DISPLAY" not in os.environ, "DISPLAY not set")
     def test_tkinter(self):