CheriBSD bugs workarounds

The correct thing to do, of course, is to fix these upstream, but that requires
understanding exactly what's wrong, and that's harder than just not tickling the
bugs.

src/pal/pal_consts.h

@@ -14,8 +14,13 @@ namespace snmalloc
    *  - using_readonly
    *  - not_using
    * model.
+   *
+   * TODO: There is a known bug in CheriBSD that means round-tripping through
+   * PROT_NONE sheds capability load and store permissions (while restoring data
+   * read/write, for added excitement).  For the moment, just force this down on
+   * CHERI.
    */
-#ifdef SNMALLOC_CHECK_CLIENT
+#if defined(SNMALLOC_CHECK_CLIENT) && !defined(__CHERI_PURE_CAPABILITY__)
   static constexpr bool PalEnforceAccess = true;
 #else
   static constexpr bool PalEnforceAccess = false;

src/pal/pal_posix.h

@@ -146,7 +146,10 @@ namespace snmalloc
 
     static void print_stack_trace()
     {
-#ifdef SNMALLOC_BACKTRACE_HEADER
+      // TODO: the backtrace mechanism does not yet work on CHERI, and causes
+      // tests which expect to be able to hook abort() to fail.  Just skip it
+      // for the moment.
+#if defined(SNMALLOC_BACKTRACE_HEADER) && !defined(__CHERI_PURE_CAPABILITY__)
       constexpr int SIZE = 1024;
       void* buffer[SIZE];
       auto nptrs = backtrace(buffer, SIZE);