Fix CI hangs on win32

After much debugging, it turns out that OpenBlas is sometimes hanging
on exit due to bad assumptions in its thread management. The upstream
discussion is at OpenMathLib/OpenBLAS#2350, but this
should fix our frequent win32 CI failures in the meantime.

deps/patches/openblas-winexit.patch

@@ -1,4 +1,4 @@
-commit 7f4a96a353c06c2fdecb0b2464800bcf8fab6fd7
+commit b16fa2e7111dc4dc1d9d987bd5dcbf89c82b9ee8
 Author: Keno Fischer <[email protected]>
 Date:   Sun Dec 29 15:08:13 2019 -0500
 
@@ -10,18 +10,18 @@ Date:   Sun Dec 29 15:08:13 2019 -0500
     of this function.
 
 diff --git a/exports/dllinit.c b/exports/dllinit.c
-index 02ff092e..56aafa95 100644
+index 4a05c0e1..88f9af65 100644
 --- a/exports/dllinit.c
 +++ b/exports/dllinit.c
-@@ -48,7 +48,10 @@ BOOL APIENTRY DllMain(HINSTANCE hInst, DWORD reason, LPVOID reserved) {
-   }
-
-   if (reason == DLL_PROCESS_DETACH) {
--    gotoblas_quit();
-+    // If the process is about to exit, don't bother releasing any resources
-+    // The kernel is much better at bulk releasing then.
-+    if (!reserved)
-+      gotoblas_quit();
-   }
-
-   return TRUE;
+@@ -50,7 +50,10 @@ BOOL APIENTRY DllMain(HINSTANCE hInst, DWORD reason, LPVOID reserved) {
+         gotoblas_init();
+         break;
+       case DLL_PROCESS_DETACH:
+-        gotoblas_quit();
++        // If the process is about to exit, don't bother releasing any resources
++        // The kernel is much better at bulk releasing then.
++        if (!reserved)
++          gotoblas_quit();
+         break;
+       case DLL_THREAD_ATTACH:
+         break;