Optimize TLS access in generated code on FreeBSD (JuliaLang#47891)

* Optimize TLS access in generated code on FreeBSD

This extends the optimization made for Linux in PR 17178 to FreeBSD. The
build and all tests pass locally with this change.

* Incorporate FreeBSD D31427 to support 12.2

src/threading.c

@@ -13,7 +13,7 @@
 // Ref https://www.uclibc.org/docs/tls.pdf
 // For variant 1 JL_ELF_TLS_INIT_SIZE is the size of the thread control block (TCB)
 // For variant 2 JL_ELF_TLS_INIT_SIZE is 0
-#ifdef _OS_LINUX_
+#if defined(_OS_LINUX_) || defined(_OS_FREEBSD_)
 # if defined(_CPU_X86_64_) || defined(_CPU_X86_)
 # define JL_ELF_TLS_VARIANT 2
 # define JL_ELF_TLS_INIT_SIZE 0
@@ -30,6 +30,11 @@
 # include <link.h>
 #endif
 
+// `ElfW` was added to FreeBSD in 12.3 but we still support 12.2
+#if defined(_OS_FREEBSD_) && !defined(ElfW)
+# define ElfW(x) __ElfN(x)
+#endif
+
 #ifdef __cplusplus
 extern "C" {
 #endif