fix #16752: threadvar now works with importcpp types; osx now uses native TLS (--tlsEmulation:off), which can be orders of magnitude faster (#16750)

* osx now uses native TLS, which can be orders of magnitude faster

* add {.cppNonPod.}

* improve test

* changelog, docs, disable part of windows test

config/nim.cfg

@@ -114,8 +114,6 @@ path="$lib/pure"
   @if bsd:
     # BSD got posix_spawn only recently, so we deactivate it for osproc:
     define:useFork
-    # at least NetBSD has problems with thread local storage:
-    tlsEmulation:on
   @elif haiku:
     gcc.options.linker = "-Wl,--as-needed -lnetwork"
     gcc.cpp.options.linker = "-Wl,--as-needed -lnetwork"
@@ -167,9 +165,13 @@ path="$lib/pure"
 
 gcc.maxerrorsimpl = "-fmax-errors=3"
 
+@if bsd:
+  # at least NetBSD has problems with thread local storage:
+  tlsEmulation:on
+@end
+
 @if macosx or freebsd or openbsd:
   cc = clang
-  tlsEmulation:on
   gcc.options.always %= "-w ${gcc.maxerrorsimpl}"
   gcc.cpp.options.always %= "-w ${gcc.maxerrorsimpl} -fpermissive"
 @elif windows: