build: enable isa-l NEON assembly on ARM macOS

macOS reports arm64 via uname -m but isa-l's Makefile.unx expects
aarch64 for the NEON/SVE assembly paths. Without this fix, only the
C baseline code is compiled, resulting in ~60% slower gzip decompression.
Pass host_cpu=aarch64 arch=aarch64 on ARM macOS to build the full
optimized library (85 objects vs 23 C-only).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: KimYannn

Makefile

@@ -38,19 +38,26 @@ GTEST_MAIN_LIB := $(DIR_GTEST)/build/lib/libgtest_main.a
 
 # On Linux: fully static binary; on macOS: static libs, dynamic system runtime
 UNAME_S := $(shell uname -s)
+UNAME_M := $(shell uname -m)
 ifeq ($(UNAME_S),Linux)
   LD_FLAGS := -static -Wl,--no-as-needed -lpthread
 else
   LD_FLAGS := -lpthread
 endif
 
+# isa-l: macOS reports arm64 but isa-l expects aarch64 for NEON assembly
+ISAL_MAKE_ARGS :=
+ifeq ($(UNAME_M),arm64)
+  ISAL_MAKE_ARGS := host_cpu=aarch64 arch=aarch64
+endif
+
 # Default target: build deps from submodules and link statically
 ${BIN_TARGET}: $(ISAL_LIB) $(LIBDEFLATE_LIB) ${OBJ}
 	$(CXX) $(OBJ) -o $@ $(ISAL_LIB) $(LIBDEFLATE_LIB) $(LD_FLAGS)
 
 # Build isa-l static library from submodule
 $(ISAL_LIB):
-	$(MAKE) -C $(DIR_ISAL) -f Makefile.unx lib
+	$(MAKE) -C $(DIR_ISAL) -f Makefile.unx lib $(ISAL_MAKE_ARGS)
 
 # Build libdeflate static library from submodule
 $(LIBDEFLATE_LIB):