loongarch64: Disable strict alignment to fix BPF program failures

CONFIG_ARCH_STRICT_ALIGN=y (default in config-6.12) causes BPF programs
(XDP, TC, tracing, etc.) to fail verification with misaligned packet
and memory access errors, breaking essential kernel functionality.

Hardware support status:
- Supported (modern CPUs): 2K2000, 2K3000, 3A5000, 3A6000, 3C5000,
                          3C6000, 3D5000
- Unsupported (legacy): 2K500, 2K1000

The current default prioritizes legacy compatibility over:
1. BPF program functionality across multiple subsystems
2. Performance on widely deployed modern hardware
3. Modern kernel features relying on unaligned accesses

Since BPF programs require unaligned access capabilities and most
LoongArch deployments use modern CPUs with hardware support, disable
CONFIG_ARCH_STRICT_ALIGN. Legacy system users can manually enable
it if needed.

Link: vincentmli/bpfire#69
Reference: https://elixir.bootlin.com/linux/v6.12/source/arch/loongarch/Kconfig#L534

Signed-off-by: Vincent Li <[email protected]>
Link: openwrt/openwrt#21121
Signed-off-by: Christian Marangi <[email protected]>

Author: vincentmli

target/linux/loongarch64/config-6.12

@@ -52,7 +52,7 @@ CONFIG_ARCH_MMAP_RND_BITS_MIN=12
 CONFIG_ARCH_SELECT_MEMORY_MODEL=y
 CONFIG_ARCH_SPARSEMEM_ENABLE=y
 CONFIG_ARCH_STACKWALK=y
-CONFIG_ARCH_STRICT_ALIGN=y
+# CONFIG_ARCH_STRICT_ALIGN is not set
 CONFIG_ARCH_SUSPEND_POSSIBLE=y
 CONFIG_ARCH_WANTS_NO_INSTR=y
 CONFIG_ARCH_WANTS_THP_SWAP=y